Transferring momentum: Novel drop protection concept for mobile devices

Abstract Dropping a tablet (or mobile phone) can be extremely costly, as loss of functionality, visible body damage, screen delamination and failure are all too familiar outcomes. This paper discusses the analysis led design of a novel passive protection concept, capable of isolating a device from the primary impact, and is also insensitive to impact angle and device dependent features. A high fidelity finite element model of an iPad Air was used to develop the BLOKTM protection concept, which utilises different grades of elastomer, optimised internal castellation geometry and a high stiffness backplate. Sensitivity studies include the influence of glass properties, screen bonding and impact angle on the robustness of the numerical predictions, whereby quantitative comparisons with experimental data in terms of metal body damage (location, size) and accelerometer data were used. Explicit finite element analysis verifies the effectiveness of decoupling the tablet from the impact loads, as resultant acceleration for unprotected versus protected was reduced by ∼76% (2152 g vs 509 g), and consistent with ∼74% reduction observed through testing (1723 g vs 447 g). For the protected tablet, simulation predicted displacements within 6%, with peak acceleration overestimated by 14%, and attributed to overestimating elastomer stiffness at full compression and its subsequent unloading. Final validation demonstrated device independence by protecting an iPad Air 2™ (with significantly different internal structure to Air™), against corner and short edge impacts. The concept developed resulted in a product to market with a mass of 165 g (∼36% tablet mass), providing protection from a 1.8 m drop onto concrete, far exceeding MIL-STD-810G requirements.

numerical predictions, whereby quantitative comparisons with experimental data in 23 terms of metal body damage (location, size) and accelerometer data were used. 24 Explicit finite element analysis verifies the effectiveness of decoupling the tablet from 25 the impact loads, as resultant acceleration for unprotected versus protected was reduced 26 by ∼76% (2152g vs 509g), and consistent with ∼74% reduction observed through 27 testing (1723g vs 447g). For the protected tablet, simulation predicted displacements 28 within 6%, with peak acceleration overestimated by 14%, and attributed to 29 overestimating elastomer stiffness at full compression and its subsequent unloading. 30 Final validation demonstrated device independence by protecting an iPad Air 2 TM (with 31 significantly different internal structure to Air TM ), against corner and short edge 32 impacts. The concept developed resulted in a product to market with a mass of 165g 33 (∼36% tablet mass), providing protection from a 1.8m drop onto concrete, far exceeding 34

INTRODUCTION 38
The mobile electronic device market is buoyant with many products available at 39 differing price levels, functionality, size and operating system. From quarterly audits 40 published by Apple TM , over 337 million unit sales worldwide have been sold since iPad 41 launch in 2010 [1]. As a result of its popularity, this paper is based upon demonstrating 42 the feasibility of a novel protection concept using an iPad Air as the reference 43 configuration for the testing and simulation campaign.
As mobile devices evolve, increased functionality and performance lead to reductions in 45 size and weight, with the Air (launched November 2013) having a 211 to 253g mass 46 saving over 1 st generation iPad (launched April 2010) [1]. As a consequence of 47 reducing mass (including metal body thickness), impact resistance and durability of the 48 protective outer glass screen and /or underlying LCD are the most damage sensitive 49 components if dropped onto a hard surface. After reviewing the literature, no papers 50 were found on the application of numerical simulation to tablets, with only a few papers 51 related to phone drops. Therefore, this review was extended to consider impact loading 52 of other electronic devices (laptops and PDAs), in addition to papers related to 53 packaging design (for protection during transit). A chronological summary of relevant 54 papers is presented in Table 1. 55 56 Table 1 -Review of electronic device drop simulation and testing, including protective 57 packaging 58 Summarising the findings from Table 1 robustness and accuracy, using LS-DYNA v971 double precision solver [2]. 136

DECONSTRUCTING AN IPAD AIR 137
Thickness data for thirty components were measured using a Creaform hand scanner 138 ( Figure 4 and Table 2). The majority of components (batteries, screen, speakers, etc) 139 are regular in shape, trivial to mesh and directly bonded to metal body. Non-structural 140 components (e.g. magnets) were represented by assigning additional mass. 141 The metal body is milled from a single piece of aluminium, has complex geometric 142 features, varying thicknesses and fillet radii. Each corner is different and requires 143 precise geometric representation and careful meshing if model accuracy is to be 144 independent of impact orientation. 145 The LCD is supported by steel (0.25mm) and plastic (1.90mm) surrounds, bonded and 146 secured to the metal body by four corner screws ( Figure 5). The outer glass is a high 147 resilience Aluminosilicate, bonded to the metal body via a 0.44mm adhesive layer. An 148 important modelling feature was the chamfered outer edge, (which provides edge 149 protection) and may under certain impact conditions, form the initial contact point and 150 directly load the outer glass. 151

MATERIAL TYPES 194
Four primary material models assigned (Table 4), with elastic properties assumed for all  195 internal components, with equivalent stiffness of Al6063-T6, as exact material types are 196 not disclosed by Apple TM . This approximation was reasonable as these components 197 could not deform, due to tight internal packaging and bonding. A rigid material was 198 used to define the impact surface and represent the inertia of the accelerometer. 199 As capturing localised damage to the outer metallic case, Johnson-Cook parameters 200 were determined from testing using coupons extracted from an iPad Air. Due to the brittle nature of glass, a Johnson-Holmquist ceramic model was required and discussed 202 in more detail in Section 3.4. 203 Table 4 -Overview of the main constitutive models assigned 204 205

OUTER GLASS CONSTITUTIVE MODEL 206
As Apple TM does not disclose materials / suppliers used, outer glass and LCD 207 specification are unknown. The glass was assumed to be aluminosilicate and based on 208 manufacturer data for Gorilla Glass TM , due to its widespread application in phones, 209 tablets and laptops [4]. The thermal and chemical treatment applied provides screen 210 resilience through a compressive preload generated in the top glass layers, which 211 minimises surface cracks opening. 212 The Johnson-Holmquist Ceramic constitutive model (MAT110), is suited for modelling 213 brittle materials, including ceramics and glass [5]. Evolution of hydrostatic pressure, Specialist dynamic tests are required to populate this material model and beyond the 228 scope of this study. Therefore, material parameters presented in Table 5 are open 229 source for silica based glass, which matches the 30.4GPa shear modulus of Gorilla 230 Glass TM [4]. In light of this approximation, this consistutive model allows for element 231 deletion (based on a user defined plastic failure strain), which could be tuned against 232 test to infer the likelihood of failure (not the mode / extent), when assessing 233 effectiveness protective concepts. 234 Table 5 -Johnson-Holquist Ceramic material properties for silica based glass [5] 235 From stability considerations, a strain to failure of 12% was required to prevent excessive 236 distortions affecting run completion (and run time considerations). Using the definition 237 of numerical accelerometers defined in Figure 7), this parameter was considered to have 238 a minor influence on the overall response, as failure strains between 6 and 20%, resulted 239 in a 4% difference in peak accelerations, Figure 6. Increasing element removal reduces 240 peak acceleration, but at the expense of (artificially) increasing case deformation (as 241 becomes weaker) and vice-versa. Strains greater than 20% resulted in element instabilities and significant reduction in calculated timestep by upto two orders of 243 magnitude (i.e. 40% element deletion reached 0.38ms before prematurely terminating and 244 were disregarded). 245 246 Figure 6 -Influence of Glass Failure Strain (element deletion) on predicted peak 247 decelerations and comparison to test (mean = 1723 ±39g) 248

INFLUENCE OF GLASS AND ADHESIVE PROPERTIES 249
For the initial drop presented in Section 2, a 1.8m drop equates to a velocity of 5.9ms -1 250 and 8.2J impact energy. Automatic tied penalty based contact algorithms (with contact 251 smoothing) were defined (as debonding not considered initially). Impact surface was 252 represented as a rigidwall. Mass proportional damping assigned, with an appropriate 253 damping factor equivalent to 10% of critical damaping [6], [7]. 254 Two numerical accelerometers were defined; at the centre of mass (referred to as 255 "Mid") and 65mm above this location as per test (referred to as "Upper"), in order to 256 assess influence of accelerometer placement ( Figure 7). Sampling rates of 10ns and a 257 termination time of 1ms allowed capturing of peak deceleration and first rebound. Nomenclature used to quantify damage is presented in Figure 8; dimensions 'a' and 'b' 269 determine major and minor axis of the dent, while 'c' represents shortening between 270 diagonally opposite corners. Resultant acceleration time histories and metal body 271 damage are presented in Figure 9 and Table 6  with the upper accelerometer predicting higher peak acceleration (2152g) than centre of 280 mass location (2094g). For both locations, accelerations are overestimated by ∼+25% 281 when compared to mean test (1723 ±39g). The initial peak is not sensitive to position 282 (t<0.5ms), whereafter, the signals reflect local dynamic effects (metal body vibration) 283 and variations in stiffness (due to packaging of internal components). Therefore, all 284 subsequent FE results will be compared to the Up(per) accelerometer, as per test setup. 285

Influence of glass and adhesive 305
To investigate how glass and adhesive flexibility / failure affects the metal body-glass 306 interface, results from the following configurations are presented in Figure 11: values (corresponding to a tied interface), resulted in higher accelerations ∼2203g 314 (+22%), as the perfect bond significantly stiffens the overall response, resulting in 315 minimal deformation and high peak accelerations. 316 Adhesive was modelled explicitly using TSHELL elements (t=0.25mm) and a perfectly 317 plastic material model. As mechanical properties for Cyanoacrylates [8] vary, two 318 extremes were chosen due to uncertainties over the adhesive used by Apple TM (Table 7). 319 320 Table 7 -Open source adhesive properties [8] Table 8 summarises commercial products (including transportation sleeves) available at 337 the time of this research, which was not exhaustive, but formed a useful starting point in 338 assessing material type and protection mechanism. 339 Dual material based designs common, as most products use a harder polycarbonate shell 340 dipped in silicone rubber to provide additional cushioning (Hard Candy, Griffin, 341 Otterbox and Gumdrop). The two sleeve cases also use polycarbonate with an 342 elastomer casing (Switcheasy) and a proprietary strain rate sensitive rubber (G-Form). 343 The Otterbox is the only product using internal foam to minimise movement. The final 344 two products (Griffin and Krakken) are substantial cases, designed to meet US-MIL-345 STD-810F (January 2000) for military and industrial applications [9]. 346 347 and temperature) and environment (moisture, sand, etc) [12]. Section "516.6 -Shock", 352 procedure IV, stipulates 26 drop tests (5 repetitions per test) from a height of 1.22m for 353 items weighing <45kg onto two inches of plywood, backed by concrete. Impact surface will affect damage, as dropping onto plywood is less severe than directly onto concrete, 355 as any localised indentation may reduce impact loading. Only the G-Form sleeve (1m), 356 Krakken (1.25m) and Griffin (1.8m) specify drop heights, with the Griffin-Survivor 357 offering the best impact protection with a mass of 349g (∼80% mass of iPad). 358

LOAD BY-PASS CONCEPT: PROOF OF CONCEPT 359
The research question posed in this paper was whether it was possible to develop a new 360 device independent protection concept sub 200grams, (∼45% mass of iPad), capable of 361 protecting from a 1.8m drop onto concrete. This represents a significant design 362 challenge due to the following design constraints: 363 • Case must represent a book in shape and appearance, completely encapsulating the 364 tablet (with case wrap around limits) 365 • Maximum 4mm wall thickness and fixed outer dimensions. 366 • Apple TM places restrictions on material types to avoid affecting wireless 367 connectivity and signal strength. 368 Existing designs absorb impact energy through elastic case deformation, typically 369 constructed from compressible foamed materials, or incompressible elastomers. The 370 main disadvantage of this approach is a limit on height protection, as the tablet forms 371 part of the primary load path (and analogous to a single spring-damper system, Figure  372 12). 373 In order to transfer momentum, the concept shown schematically in Figure 12 and in 377 more detail in Figure 13, aims to decouple the device from the impact loads by utilising 378 a high in-plane stiffness back plate insert, whose dimensions are larger than the tablet. 379 The device would be encapsulated by an inner (softer) elastomer, providing localised 380 load alleviation and allowing relative motion between device and case (and analogous to 381 two "springs in parallel"). Decoupling the device will permit decelerations over a 382 longer duration, thereby reducing loading and offering increasing protection across a 383 wider range of drop heights. Key to success is allowing controlled deformation of the 384 inner elastomer. If the elastomer densifies, the tablet would become part of the primary 385 load path, thereby increasing accelerations experienced. 386 CPUs, Intel E5 -2660 CPU) to complete 0.6ms (sufficient to capture the initial peak 393 deceleration). These materials were chosen as commonly available from elastomer 394 suppliers, in addition to hyperelastic material coefficients available in the literature 395 (Table 9). Two different sets of coefficients were used for comparable hardness 396 rubbers, as parameters for the Yeoh strain energy function [13], (a third degree 397 polynomial allowing the shear modulus to vary with deformation), hereafter referred to 398 as "Yeoh52A", as coefficients were reported to be stable under large deformations [14]. 399 Table 9 -Initial material properties (rubber and polycarbonate) [14]  • Presence of the cover delays the time the tablet experiences a peak acceleration, 414 which is caused by relative motion between case and tablet. The peak accelerations 415 occur after ∼0.25ms and ∼0.40ms for unprotected and protected tablet respectively. 416 • When the inner elastomer 'bottoms out' (i.e. reaches a maximum compression of 417 1.6mm), the solid inner elastomer surround offers very high resistance to increased 418 compression, which is reflected by only a 9% reduction in peak acceleration. 419 • The only practical means of reducing peak acceleration is through careful elastomer 420 design. Inclusion of voids in the inner elastomer surround needs to be optimised to 421 promote a controlled 'flow' of rubber (i.e. deformation through shear) in order to 422 avoid 'bottoming out' and subsequent (infinite) stiffening.

DESIGN VARIANTS FOR ELASTOMERIC INNER 424
A systematic exploration was undertaken, which considered the inclusion of recesses in 425 the inner elastomer, either internally or externally, together with different orientations 426 (i.e. along hoop and radial directions). Based on manufacturing considerations (which 427 includes minimum thickness constraints for single pass injection moulding (to ensure 428 quality / repeatability), the aim was to reduce inner elastomer mass by up to 30%, 429 resulting in six variants analysed. All simulations utilised the initial material properties 430 in Table 9 and took approximately 10-12hrs on 16CPUs (MPP cluster, Intel E5-2660). 431 Key findings are summarised in Table 10, together with normalised resultant 432 accelerations in Figure 15. 433 In general, the impact sequence is divided into four stages: 434 1. Initial Impact -all contacts (internal and with impacting surface) established. 435 2. Stage I -Inner elastomer, which is more flexible than the outer cover deforms in 436 shear to fill internal voids. Once the high stiffness back plate makes contact with 437 the impacting surface, the back plate becomes the primary load path. 438 3. Stage II -represents a dwell period, where device moves relative to back plate and 439 is brought to a complete rest by stretching of the inner elastomer. Back plate helps 440 outer cover retain its shape and more importantly, tablet encapsulation.

DEVELOPMENT OF THE "BLOK TM " PROTOTYPE 467
The internal geometry of TPE (Thermoplastic Elastomer) voids were optimised by 468 considering usability, industrial design and design for manufacture ( Figure 16). 469 Optimised configuration consists of localised, angled castellations, which were 470 discontinuous at each corner to prevent the inner elastomer bottoming out under a direct 471 corner impact. Along long and short edges, the castellations were orientated normal to 472 the device to provide optimal (distributed) protection from edge impacts. The outer was 473 an 80 Shore A TPE (ThermoPlastic Elastomer), as an increase in stiffness was required 474 to ensure tablet retention, with a less stiff TPE for the inner elastomer.  The kinematics of the tablet are in close agreement in terms of rise time and absolute 507 magnitude, with predicted peak displacement of 8.98 vs 8.45mm (+6%) and close 508 agreement up to 1.8ms for resultant velocity, whereafter the simulation diverges from 509 test. In terms of resultant accelerations, LSDYNA predicts the rise and fall well, but 510 overestimates the peak deceleration by +14% (509g vs 447g at t = 1.8ms).
The deviations after 1.8ms suggests LSDYNA overestimates inner elastomer stiffness 512 (at full compression) and its subsequent unloading. However, the simulation verifies 513 the effectiveness of this proposed concept, as the (predicted) resultant acceleration for 514 an unprotected and protected tablet has been reduced by ∼76% (2152g vs 509g). This 515 observation is also consistent with test: 1723g (unprotected) vs 447g (protected); a 74% 516 reduction. Overall, there is very good agreement to test. 517

FINAL VALIDATION CASE -APPLICATION TO IPAD AIR 2 518
During the development of this research, the iPad Air 2 was released, which internally 519 contained many differences to the Air (Table 11)  lack of LCD support as it is directly bonded to glass screen), resulting in the glass 531 screen being more prone to cracking / shattering. 532 533 Table 11 -Internal differences between iPad Air and Air 2 [17], [18]  • Glass-metal body bonding (and failure) are key parameters that can affect the 555 overall table response, which would require additional characterisation to align 556 simulation to test.
• Decoupling the tablet from the impact loads has significant advantages and the 558 potential to provided increased protection to any device (or mobile phone) from 559 even greater drop heights, as the primary load path is through a high stiffness 560 backing plate (not the device). 561 • The developed model verifies the effectiveness of this concept, as the (predicted) 562 resultant acceleration for an unprotected and protected tablet was reduced by ∼76% 563 (2152g vs 509g), and consistent to test: 1723g (unprotected) vs 447g (protected); a 564 74% reduction. Displacements agree within 6% and peak acceleration was 565 overestimated by 14% and attributed to LSDYNA overestimating elastomer 566 stiffness at full compression and its subsequent unloading. 567

FUNDING 583
This research did not receive any specific grant from funding agencies in the public, 584 commercial, or not-for-profit sectors. 585 Table 1 -Review of electronic device drop simulation and testing, including protective   590   packaging   591   Author  Date  Conclusions Suhir [19] 1994 Analysed the dynamic response of a rectangular plate (e.g. liquid crystal display) and demonstrated that plate thickness and clamping could improve failure strength.
Nagaraj [20] 1997 Applied numerical simulation to determine PCB integrity by assessing vibrational mode shapes and identifying maximum displacements and potential impact points within case.
Goyal [21] [22] 1999 2000 Identified thin-walled clamshell phone cases may not have sufficient structural rigidity to resist impact loading. In 2000, battery integrity was assessed, which identified a problem with manufacturing. Author proposed an automated drop testing method to improve repeatability.
Low [23] 2001 Modelling, simulation and test correlation for 0.6, 0.8 and 1.2m drop heights of hi-fi products onto concrete. Numerically, difficulties encountered with accuracy of material properties and small time steps, resulting in a sub modelling approach recommended for damage assessment of internal components.
Lim [24] 2002 Varied impact height (0.55 to 1m) and angle for drop testing an electronic pager, where LCD represented with solid elements (including glass), equivalenced to membrane elements representing the lens (to extract surface strains). Correlation with force transducers, accelerometers and strain gauges showed importance of minimising idealisation errors, with vibration of lens / LCD identified as cause of failure due to cyclic strains.
Seah [25] 2002 Supported observation that impact angle and device specific design are important parameters, strongly influencing PCB failure, where shock induced vibration results in differential bending and extreme cyclic stress, resulting in crack propagation / separation under low cycle fatigue.

Lim [26] [27] 2002
2003 Impact angle sensitivity to phone / PDA devices, identifying maximum forces for direct vertical / horizontal impacts to be 2 to 5 times higher than for oblique impacts. Loading heavily influenced by dimensions and device composition. If internal components are in direct contact with outer case, loading directly transmitted and is severely damaging.
Low [28] 2003 Developed simplified models for hi-fi, hard disk drive and irons to predict transient impact response through equivalent spring-mass systems.
Influence of mass on the benefits of providing a cushioning buffer (Expanded polyethylene) to reduce shock loading was investigated, which controlled spring back of the buffer material.
Low [29] 2004 Utilised a sub-modelling approach for 0.5m drop impact analysis of a 29" TV, where the local model concentrated on the detailed analysis of four tube attachment screws. Increasing the radius of curvature at the impact point is beneficial in reducing stress concentrations.
Lye [30] 2004 Applied Genetic algorithms to optimise protective packaging buffers using bio-degradable materials (paper pulp and starch). Testing conducted on six configurations resulting in agreement within 12%.
Yi [31] 2005 Applied a Design of Experiments to optimise Expanded polystyrene monitor packaging, using an automated optimisation framework reducing design cycle from weeks to days. Approach is complex and worked well for homogeneous packaging, which could be applied to tablet protection.
Wang [32] 2005 Reviews impact response of different devices including styrofoam modelling for a TV and a PCB with an interference fit. Conclusions general, with recommendations made regarding importance of simplify material geometry (due to long CPU times / mesh distortions), definition of contact and material model accuracy.
Tan [33] 2005 Investigated simplified and detailed drop test modelling of printed circuit board (PCB), integrated circuits (IC) and interconnecting solder balls to determine PCB deflection and solder stresses. Large variation in predicted stress observed due to complex interaction between components, necessitating detailed solid element models for correlation.
Cadge [34] 2006 Considered failure of joints through a 1m drop of an optical computer mouse. Approach adopted included sub-modelling, whereby system level time histories were extracted and applied to a detailed joint model using ABAQUS. Modelling approach identifies a location for possible joint failure / detail stresses, but no experimental validation provided.
Shan [35] 2007 Developed an analytical model for a rigid body impacting multiple times onto a viscoelastic surface. Subsequent impacts can occur at multiple case locations and may be more severe (in terms of contact force), due to random angles of impact.
Pan [36] 2007 Simulation and test correlation for 0. 5       • Inner elastomer flows into space/ void created between Inner and Outer cover, minimising case splaying (-35% reduction in peak acceleration). t = 0ms t=1.0ms • Circumferential grooves increase risk of case splaying due to reduced second moment of area of case along hoop direction (-5% decrease in peak acceleration) t = 0ms t=0.5ms • -5% reduction in peak acceleration, due bottoming out of inner elastomer (similar issues as Variant B). t = 0ms t=0.9ms • Corner lug reduces internal space available for inner elastomer, resulting in elastomer bottoming out quickly (+17% increase in peak acceleration). t=0.6ms • Similar issue as with Variant D -Inner lateral grooves beneficial, but no net benefit over baseline (+2.3% increase in peak acceleration). • -29% reduction in peak acceleration, as absence of corner lug allows inner elastomer to deform in shear.