Mohammed Musharaf Hussain. Additive manufacturing, systems, imaging, sustainability.
Parts under load, drawn honestly.
Four projects where something has to carry load: a nozzle under thrust, a relay in humid air, a heart under analysis, materials under net zero. What broke is part of the story.
Four projects and a log.
001 /Rocket nozzle for additive manufacturingDissertation
A de Laval nozzle sized in Python from Merlin 1D baseline inputs, modelled in SolidWorks, then run through an ANSYS laser powder bed thermal simulation and CFD. Designed and simulated, not printed. The structural analysis never finished, and the write-up says so.
002 /Q-relay redesignSiemens partnership
In humid conditions, phenolic insulators act as an electrolyte and let silver migrate. Unfilled PEI, chosen from an Ashby chart, costs under £0.12 a piece with CNC machining. Case study +Case study +
003 /Atrial segmentationMedical imaging
Right atrium from LGE-MRI, painted by hand in 3D Slicer and scored in MATLAB. A human baseline rather than machine learning, and it under-segments badly.
004 /Materials for net zeroSustainability report
Fuel cells, thermoelectrics and magnet motors set against UK net zero 2050, told as candidate material swaps: PbTe to GeTe, NdFeB to SmFe₁₂.
005 /LogOngoing
6 Oct 2026. Hero and index prototype built. Case studies come next, starting with the nozzle.
Case study 002 / Siemens partnership
Q-relay redesign
Siemens asked a team of five to work out why the operating arms and adjustment cards in its Q-relay were causing trouble, and to propose a fix. In humid air the phenolic insulating material acts as an electrolyte, which lets silver migrate between nearby metal paths. That can cause short circuits and wrong-side failures.
The Q-relay, photographed. Scroll and it turns to type.
Choosing the material Ashby chart
Redrawn schematic. Bubble positions are illustrative, and the original chart was made in Ansys Granta EduPack. The diagonal marks the index M1, cost per unit mass divided by tensile strength. Lower is better, so bubbles above the line drop out. Three survive: PEI (unfilled), PET (15% glass fibre, flame retarded) and polyester SMC (15% glass fibre, V-0).
Decision matrix weight x score
PET scored highest, but it is not recyclable. Siemens treats sustainability as a priority, so the team recommended unfilled PEI instead and accepted the lower score. Phenolic materials were ruled out for the silver migration problem, and fibre-filled options were disfavoured because loose fibres from blanking were one of the existing faults.
Testing the part hand calculation and FEA
Hand calculations treat the card as a 47 x 7 x 1.5 mm cuboid, as the report does. The FEA is on CAD of the real geometry. The report states an average factor of safety above 1, excluding the torsion test.
Cost per component limit £0.50
Unfilled PEI sheet ~£0.058
CNC machining ~£0.0525
Total with CNC ~£0.115
In-house extrusion, rejected ~£0.173
Water gain, PEI / PET / SMC 20.5 / 6.81 / 12.0 mg
Not done yet what I would do next
Blanking costs could not be calculated for lack of data, so CNC stands in. The heated-blanking idea comes from a study of magnesium alloy AZ31B (Fazily et al., 2019) and has not been tested on PEI. A rubber sealant would help against moisture but could undermine recyclability. PEI also absorbs the most water of the three, within the limit but with the least headroom.
The numbers, including the unflattering ones.
0.58
Fig. 3. Dice score for the atrial segmentation. The orange arc is this work at 0.5766. The bar on the scale marks deep learning and expert agreement, 0.91 to 0.94.
001 / Nozzle Python and CFD
Throat velocity, calculated 1173.5 m/s
Exit velocity, calculated 2765.3 m/s
Exit velocity, CFD ~2350 m/s
Isp, CFD against Merlin 1D 239.6 / 282 s
002 / Q-relay per piece
PEI sheet ~£0.058
CNC machining ~£0.0525
Cost limit £0.50
003 / Atrial segmentation right atrium
Precision 0.8005
Dice 0.5766
False negatives 167,321
004 / Net zero g CO₂ per kWh
Coal against solar 943 / 25
Before you ask.
Q.001 /Was the nozzle printed?
No. It was designed in CAD and simulated for additive manufacturing. A physical print and testing are listed as future work.
Q.002 /Does it match the Merlin 1D?
No. Merlin is a baseline, not a validated match. The throat is 0.267 m against about 0.226 m, and the CFD Isp is 239.6 s against 282 s.
Q.003 /What broke?
The structural analysis. Repeated ANSYS crashes, a corrupted Workbench file and an unresolved error code stopped the thermal-to-structural and CFD-to-structural coupling.
Q.004 /Is the atrial work machine learning?
No. It is threshold-based and hand-painted segmentation of the right atrium, scored against a gold standard. Deep learning appears only as a benchmark.
Q.005 /Was the nozzle a solo project?
No. It was one work stream in a group project to design a fully reusable rocket with additive manufacturing. Mine was the nozzle: design, thermal management, material selection and process choice.