Portfolio

Nishita Philip

Reliability engineer with a materials engineering background

I plan and run hardware test campaigns, including mechanical and environmental testing. When something fails, I trace it to the material and process behind it and recommend a design or process change.

fortisfille@gmail.com linkedin.com/in/nishita-phillip Mountain View, CA
Nishita Philip
About me

I'm a reliability engineer who thinks like a materials scientist. I design the test, run it, and when something breaks I don't stop at the crack. I trace it to the material or process behind it and bring back a fix the design team can act on.

I've done this across industrial manufacturing, AR/VR wearables and flagship consumer electronics.

I do my best work where the standard answer isn't enough: a failure nobody can reproduce, a test method that doesn't exist yet, or a result so borderline that the ship decision depends on reading it right.

Next, I'm bringing that rigor to larger, safety-critical electromechanical systems, where a failure costs far more than a warranty claim.

Off the clock: video games, dancing, and learning about new cultures.

Focus
Reliability testing, failure analysis, materials
Methods
DVP&R, DFMEA, HALT/HASS, DIC, SEM/EDS
Where I'm headed
  • Safety-critical systemsBring consumer-electronics test rigor to electromechanical assemblies where failure matters most.
  • Design-to-field reliabilityOwn the reliability plan from first prototype to field data.
  • Product decisionsTurn reliability risk into clear calls on what gets built and when it ships.
Education

Materials Science and Engineering

M.Eng., University of California, Berkeley2024
B.Tech., Indian Institute of Technology Jammu2023

Other projects: paper-based biosensor for rapid bacterial detection (cut UTI diagnosis time by 75%), rare-earth magnetostrictive alloy design, and a 2022 publication on microstructure evolution in alloy design.

How I work
01

Plan

Turn use conditions into a test matrix with clear pass criteria.

02

Test

Run instrumented tests and record how the hardware behaves.

03

Analyze

Find the failure mode and its root cause from the data and the failed parts.

04

Feed back

Report the root cause and a specific design, material or process change.

Selected work

Case studies

Industry projects describe methods and outcomes only. Product details and data from Apple and Meta are confidential.

Qualifying conformal coatings for wearable flex circuits

Meta Reality Labs · Materials Lab
Challenge

Pick one of eight conformal coatings for the flex circuits in a wearable. The coating has to keep sweat and moisture off the copper while the circuit bends, and the vendors' datasheets weren't comparable.

Test coupon: an unpopulated flexible circuit strip with three copper traces, coated in the middle and left bare at the probe pads on each end probe pads (bare) coating 25–75 µm polyimide pad-to-pad resistance copper
Each sample is an unpopulated flex strip, coated except for the probe pads so resistance can be measured after every step.
Bend test: the same coated strip is bent 180 degrees over rods of 4, 3, 2 and 1.5 millimetres; the coating cracks only on the smallest rod no cracksno cracksno cracks cracks 4 mm3 mm design2 mm1.5 mm
Critical radius is the smallest rod that leaves the coating uncracked. Leg 2 repeats this after the hot-car cycle to catch embrittlement.
Test grid · 8 vendors × 8 legs × 20 samples = 1,280 samples
MechanicalThermalHumidityChemical
Coating properties
1Monotonic bendsingle stepStatic flex limit, brittle coating
2Hot car, 25 → 85 °C daily, 30 daysMonotonic bendHeat damage, then embrittlement
Environment controls
385 °C / 85 %RH, 1,000 hsingle stepBarrier failure, hydrolysis, blisters
4Sweat, lotion, IPA, 72 hsingle stepSwelling, softening, solvent attack
Flex then environment
5Cyclic bendHumidityBend fatigue → moisture corrosion
6Cyclic bendChemicalBend fatigue → sweat attack
7TorsionHumidityTwist damage → moisture corrosion
8TorsionChemicalTwist damage → sweat attack
What I did
  • Designed an eight-leg test grid. Four legs apply one stress, and four apply bending or twisting first, then humidity or sweat.
  • Checked every sample after every step for coating damage and for a change in electrical resistance.
  • Used identical test strips, per-sample baselines and randomized chamber positions, so the coating was the only variable.
Key findings
  • Vendor C passed 157 of 160 samples, the best of the eight.
  • The combined legs separated the vendors. C led F by 2 passes on single-stress legs and by 10 on combined legs.
  • Vendor E failed mainly when flexed and then exposed to sweat, which single-stress testing would have missed.
Samples passing, out of 160

Representative data. Vendors are anonymized.

Result

Recommended vendor C, conditional on qualifying production lots and adding an adhesion check on incoming lots, with vendor F as the second source.

Conformal coatingsFlexible circuitsVendor qualificationTest designHumidity / 85-85Chemical exposureFailure analysis

Cutting scrap on heat-treated bearing rings

SKF India
Challenge

Too many bearing rings were coming out of heat treatment distorted and had to be scrapped. I was asked to find out why and fix the process.

Where the problem showed up
  1. Raw steelBearing steel bar or tube.
  2. FormingForging or tube cutting, then soft turning to near-net shape.
  3. Heat treatmentFour furnaces with preheat, austenitizing, quench and tempering zones. My focus.
  4. FinishingGrinding to final dimensions.
  5. InspectionQuality checks. Distorted rings were scrapped here.
What I did
  • Tested four possible causes one at a time: steel and upstream processing, furnace, furnace zone, and cooling.
  • Measured about 80 rings per trial on a dial gauge, before and after heat treatment.
Schematic of a nominal bearing ring compared with an oval and a grown ring nominal ovality growth
The failure modes we classified, exaggerated for clarity. Dashed line is the nominal ring.
Top view of rings on a furnace belt: on the undersized belt rings pile up and touch at one point; on the resized belt they travel evenly spaced before · undersized beltafter · resized belt rings pile up and touch→ direction of travel
Illustration of the belt finding. Rings that piled up pressed on each other while hot and soft, and distorted. A wider belt kept them apart.
Key findings
  • One furnace's belt was too narrow, so hot rings piled up and pressed on each other.
  • Ring stacking and loading, and one steel and pre-processing route, made the distortion worse.
Result
−70%

Distorted-ring scrap after the plant replaced the belt, changed ring loading and rerouted the problem material. Cost and emissions per good ring dropped with it.

Root-cause analysisHeat treatmentDesign of experimentsDimensional metrologyBearing steelScrap reduction

Catching hidden bond failures in drop testing

Consumer electronics reliability
Challenge

Find out whether a structural adhesive bond in a handheld device holds up to bending, pressure and drops, including failures that leave no trace after the test.

Three tests
Four-point bend: a panel, adhesive and housing stack rests on two supports while two loading noses press down between them paneladhesivehousing cyclic, and to peak load

Four-point bend

Cyclic and to peak load, for fatigue life and strength.

Compressive push-out test: the assembly is held housing side up, and a plunger presses down through a hole in the housing onto the bonded panel until the adhesive separates force housing bonded panel plunger presses down through the housing

Push-out

A compressive test. Gives peak separation force, and whether the bond failed on the panel or the housing side.

Drop test: a device falls corner first onto a floor of load-cell tiles while high-speed cameras record it for digital image correlation load-cell tiles · high-speed DIC

Drop

Corner and edge drops at the angles simulation flagged as worst.

What I did
  • Ran four-point bend and compressive push-out tests to measure bond strength and where it fails.
  • Dropped devices onto a floor of load-cell tiles that measured the impact force at each corner and edge.
  • Filmed every drop with stereo high-speed cameras and fit the CAD model to the DIC data to measure panel-to-housing separation.
Illustration of panel-to-housing gap over time during a drop: the gap is zero before impact, spikes open at impact, then closes back to zero, so post-test inspection shows nothing impact bond opens post-test inspection: no gap time panel–housing gap
Illustration. The bond opens at impact and closes again, leaving no visible gap.
Outcome

Built a combined method that measures the bond in the moment of impact, not just after the test. Short-lived separations that close again leave nothing for post-test inspection to find; high-speed DIC can catch them, so they can be fixed before they grow into field failures.

Structural adhesivesFour-point bendPush-out testDrop testingLoad-cell floorHigh-speed DICSimulation-guided testing

Quantifying what biases a phone's impact orientation in random drops

Consumer electronics reliability
Challenge

A phone's design can bias it toward landing in particular orientations, which puts those spots under the most stress. We compared the factors behind that bias: weight distribution, starting orientation, and the camera turret (the raised platform that holds the cameras), including its position, its size and whether the phone has one at all. Seeing a bias reliably takes a very large number of drops, more than a conventional drop setup can deliver in reasonable time.

A phone tumbling inside a rotating drum while a stereo pair of high-speed cameras, mounted side by side on one rail, records every drop fixed RPMstereo pair(high-speed)
The tumbler lifts and drops the phone at random on every revolution.
The phone's CAD model is fitted to the DIC measurement of the falling phone, giving its three-dimensional orientation at impact xyz measured surface (DIC) fitted CAD
The fitted CAD model gives the phone's 3D orientation at impact.
What I did
  • Developed a new test method around a tumbler, a rotating drum that drops the phone repeatedly at random.
  • Recorded every drop with stereo high-speed cameras and fit the phone's CAD model to the DIC data to get its orientation at impact.
  • Automated most of the process, so each revolution produced drops with no manual reset.
Illustration comparing the same stretch of time: a conventional drop study spends most of it resetting between drops, while the tumbler produces drops continuously manualtumbler … set up, drop, inspect, reset drops captured continuously, every revolution
Illustration. Orange ticks are drops. The manual study stops to reset before every drop.
Result

Because the method captured so many drops, we could see which orientations the phone favored and how weight distribution, starting orientation and the camera turret shifted that bias. About 80 drops took a small fraction of the time of a conventional drop study, and the results showed which orientations to prioritize in controlled drop tests.

Test method developmentImpact orientationTumble testingStereo DICHigh-speed imagingCAD fittingTest automation

Tailoring carbon fiber layouts for a wearable frame

Meta Reality Labs
Challenge

Some parts of a wearable frame have to stay stiff and others have to flex. The goal was carbon fiber layouts with the right stiffness in each region.

Illustration of an eyewear frame preform made from one continuous carbon tow that never crosses itself and turns only in rounded loops: five lanes on each temple, five above and four below each lens, and three wider-spaced lanes across the bridge start end temple · 5 lanes temple · 5 lanes bridge 3 lanes rim · 5 top / 4 bottom rim · 5 top / 4 bottom
Illustration of one unbroken carbon tow. Temples and rims have tightly spaced lanes for stiffness, and the bridge has three wider-spaced lanes so it can flex.
Process
  1. DesignFiber layouts drawn in ZSK's design software, varying tow path and density by region.
  2. FixtureFabric loaded into my adjustable magnetic hoop, sized to the sheet.
  3. StitchCarbon tows stitched onto polymer base fabrics on the ZSK machine.
  4. MoldStitched preforms compression molded into composite parts.
  5. TestBend, flex cycling, thermal, hardness and thermal conductivity, with SEM/EDS on failures.
What I did
  • Compared 3K, 6K and 12K tows, 0°/45°/90° orientations, and different tow spacing, width and curvature.
  • Tested molded parts with bend, flex-cycling, thermal, hardness and thermal conductivity tests.
  • Designed a 3D-printed, magnetic fixture in SolidWorks to hold fabrics of any size on the stitching machine.
Illustration of the adjustable magnetic hoop: each side is a pair of bars that clamp the fabric between them, held by magnets along their length. Inside the frame the fabric is taut and its weave runs straight; outside the bars the loose excess is wrinkled and wavy. taut slides to resize slack outside: wrinkled one side, in section top bar bottom bar fabric clamped between, magnets pull bars together
Illustration, not the actual part. Each side is two bars that clamp the fabric between them, held together by magnets along their length, so the fabric inside the frame stays taut and its weave runs straight, while the slack outside wrinkles. Sliding a side resizes the opening.
Key findings
  • Tow spacing, width and size set local stiffness, so one part can have several stiffness zones.
  • Tight and loose curves change stiffness without adding or removing fiber.
  • The fixture made changeovers faster and let the team use every available fabric on the machine.
Result

Guidelines for tuning stiffness by region, which fed into material choices and design revisions for the frame program.

CFRPTailored fiber placementZSKCompression moldingSolidWorks3D printingFlex cyclingThermal conductivitySEM/EDS

Validation plan for an outdoor robot actuator

Design exercise · hypothetical
Challenge

A robot joint actuator has to run for months on construction sites, in rain, mortar dust, −20 to +50 °C and knocks. How would I show it is reliable before it ships to a growing fleet, while the design is still changing quickly? My industry test plans are confidential, so this hypothetical plan shows my approach. All numbers are illustrative assumptions.

Schematic of a robot joint actuator with encoder, motor, gearbox, output bearing, shaft seal and cable gland, exposed to rain, dust and mortar, temperature cycles, vibration and impact rain dust, mortar impact vibration temperature cycles encoder motor gearbox output bearing shaft seal cable gland
Hypothetical joint actuator and the site conditions it has to survive.
Top risks first (DFMEA excerpt, highest priority at the top)
PartFailure modeSite causeTestWhere
Shaft sealWater, mortar or dust gets inRain and hosing, abrasive cement grit, thermal pumping (a hot joint cools and sucks in water)IPX6 jets (IPX9K if robots are pressure-washed) and IP6X dust; run hot, then cold-spray; run the seal in real cement dust, not just talc. Repeat all after life cyclingIn-house
GearboxOverload damage, wearCollisions, dropped bricks, gritInstrumented collision and overload torque test, then endurance rigIn-house
Gearbox, coldStiff grease overloads the motor−20 °C morningsCold soak at −20 °C, then measure start-up torque and currentIn-house
MotorOverheating, torque deratingSummer sun on a dark housing, continuous dutyRun the duty cycle at +50 °C with solar load, logging winding temperatureExternal lab
Seals, plasticsSwelling, crackingWet mortar, pH 12–13; UVSoak samples in cement slurry, check hardness and size; UV exposure (ISO 4892)In-house
Cable, connectorFretting, cable fatigueTransport vibration, flexingFlex cycling, including at −20 °C when cables stiffen; random vibration to a truck transport profile (IEC 60068-2-64)External lab
EncoderCondensation, signal lossTemperature and humidity swingsPowered damp-heat cycling (IEC 60068-2-30)External lab
Output bearingDents, fatigueImpacts on the armSupplier ratings, checked against impact loads measured on a site robotSupplier data
Test plan
  1. Measure the real dutyInstrument a robot on site with strain gauges, accelerometers, motor current and temperature. Every rig profile comes from this load spectrum, not from guesses.
  2. Rank risksUse the DFMEA and past field failures to test the top few risks, not everything.
  3. Accelerate with careCut idle time and raise load using known life exponents (bearing life ∝ load−3). Keep temperature inside grease and seal limits so the failure mode doesn't change.
  4. Fail, fix, re-testRun 3–5 units per design revision to failure and check each failure matches the field. After a fix, re-run that test plus a short regression set (IP after cycling, cold start).
  5. Fleet as test bedLog failures per bricks laid and per hour, by design revision, to track reliability growth. Trend motor current and temperature to catch wear early.
In-house endurance rig: the actuator is clamped to a bench and swings a weighted arm through its duty cycle, while a small data logger records motor current, temperature and cycle count actuator weighted arm, duty cycle 24/7 logger current, temp, cycles
Illustration of an in-house endurance rig. It runs the real duty cycle around the clock and logs the signals that drift before a failure.
Why test to failure, and what it can't prove

If a robot has six joints in series, each needs about 98% reliability over the target life for the whole robot to reach 90%. Demonstrating that at 90% confidence with a zero-failure test takes 29 units run to twice their life, and a pass says nothing about how or when the joint would fail. With a design that changes every few months, I would rather run a few units to failure. That shows the failure mechanisms and the margin quickly, so each revision gets better. A few failures can't demonstrate 98% on their own, so the formal number comes from fleet hours, which grow with every robot deployed.

ApproachUnitsWhat you learn
Zero-failure demonstration, 98% per joint, 2× life29A demonstrated number, but pass or fail only, and slow to repeat after each design change
Test to failure per revision, then fleet data3–5 + fleetHow it fails and the margin now; a life estimate that starts wide and tightens with every brick laid

Assumes a series system with equal allocation (0.91/6 ≈ 0.983) and Weibull β ≈ 2 for wear-out. Units for a zero-failure test = ln(1 − C) ÷ (kβ · ln R). With only 3–5 failures, β is uncertain, so I would anchor it with supplier and field data (Weibayes).

What the plan delivers

A plan that keeps pace with the design: test profiles taken from real site loads, in-house rigs for quick iteration, external labs where they add the most, and early failure data from a few units. Each test traces back to a failure a construction site can cause, and the deployed robots turn the life estimate into a demonstrated number after launch.

DFMEAAccelerated life testingWeibullReliability growthVibration and shockIP testingTest rig designField data

Designing a high-entropy alloy for nuclear reactors

UC Berkeley capstone, with Lawrence Livermore National Laboratory
Challenge

Design an alloy for reactor structures that stays strong at 300 °C and 150 bar, resists corrosion and absorbs few neutrons. Team project advised by Prof. Peter Hosemann.

Capstone workflow in four stages: CALPHAD modeling, arc melting, heat treatment, and testing for oxidation, corrosion and helium-ion irradiation CALPHADphase model Arc melting5 elements Heat treatmentmicrostructure Exposure testsheat · ions
We predicted the phases, made the alloy, set its microstructure with heat treatment, then exposed it to heat, corrosion and helium-ion irradiation.
What I did
  • Chose Al, Ti, V, Cr and Mo using a neutron-absorption figure of merit, and confirmed with CALPHAD modeling that they form a single BCC phase.
  • Arc-melted the alloy and characterized it with SEM, EDS, fractography and Vickers hardness.
CALPHAD phase diagram of VMoCrTi with aluminum, showing a single-phase BCC region across aluminum fractions
CALPHAD predicted a single-phase BCC solid solution.
Measured composition by EDS, in atomic percent: Ti 21.3, Cr 20.8, V 20.1, Mo 19.9, Al 18.0, against the 20 percent equiatomic target 20% equiatomic target TiTi: 21.3 at.%21.3CrCr: 20.8 at.%20.8VV: 20.1 at.%20.1MoMo: 19.9 at.%19.9AlAl: 18.0 at.%18.016%18%20%22%
EDS: every element within about 2 at.% of the equiatomic target.
Backscattered SEM image at 200x showing a dense dendritic microstructure
SEM (BSE, 200×): dendritic microstructure.
SEM fractograph showing flat cleavage facets
Cleavage facets, typical of brittle fracture.
SEM fractograph with a grain boundary fracture region outlined
Grain boundary fracture (outlined).
Key findings
  • The composition landed within 2 at.% of equiatomic, as designed.
  • Hardness was 330 ± 34 HV, with a dendritic microstructure.
  • Fracture was mostly brittle, which we traced to oxide layers between grains.
Result

Later oxidation, corrosion and helium-ion irradiation tests identified the alloy as a candidate corrosion- and wear-resistant coating for reactors. Planned next steps were a remelt, new cooling rates and a Mo-free variant.

Alloy designCALPHAD / Thermo-CalcArc meltingSEM / EDSFractographyVickers hardnessNuclear materials
Toolkit

Capabilities

Mechanical testing

UTM tensile, fatigue, fracture toughness, four-point bend, ball-push, ring-on-ring; drop tower, horizontal launcher, tumble; hardness.

Environmental and accelerated

HALT/HASS, thermal cycling, IPX7, humidity, salt fog, chemical exposure, corrosion, high-temperature oxidation.

Data and instrumentation

Python, JMP, LabVIEW; Kistler DAQ; Phantom and Photron high-speed imaging; DIC with GOM/ZEISS Inspect; SolidWorks; Thermo-Calc.

Characterization

SEM (SE/BSE, fractography), EDS, Keyence microscopy, metallography; XRD, XPS, FTIR, UV-Vis; DSC, DMA, TGA.