Die-Sinking EDM: The R0.05 Corner, the 30% Wear Tax, and the 0.07 mm That Eats Three-Quarters of the Hours
A 100 × 80 × 20 mm pocket in H13 tool steel, hardened to 50 HRC, arrives at the EDM bay from the milling department with 0.62 mm of stock on the walls and floor. The toolmaker has four electrodes on the bench: two graphite — coarse for roughing, finer for semi-finish — and two copper for the finishing passes.
The first 0.55 mm of that stock comes off in about 96 minutes. The last 0.07 mm takes 279 minutes. Same machine, same dielectric, same night shift — a drop from 300 mm³/min in roughing to 2 mm³/min in the final pass, bought deliberately because the drawing says Ra 0.4 µm and ±0.01 mm, and nothing about the process is linear on the way there.
That is the economics of die-sinking EDM in one sentence: you pay for the last few micrometres with hours, and you pay for the electrode twice — once when you mill it, and again when the spark eats it. Everything else in this article is the arithmetic behind those two sentences.
1. One spark, 20,000 °C, and a crater a few micrometres deep
Sinker EDM removes material by controlled electrical breakdown across a gap filled with hydrocarbon oil. The tool — a shaped graphite or copper electrode — never touches the workpiece. The dielectric breaks down, a plasma channel forms, and the temperature at the crater spot reaches roughly 8,000 to 20,000 °C for the duration of the pulse. Steel melts at about 1,540 °C and boils at 2,860 °C. The plasma does not care that the steel is 50 HRC.
The energy delivered in one discharge is the integral of gap voltage times current over the pulse-on time. For a rectangular generator pulse it collapses to a product:
with V_d the discharge voltage once the channel is established (20–30 V, not the 80–120 V open-circuit value), I_p the peak current, and t_{on} the pulse duration. A roughing pulse at 40 A for 200 µs against a 25 V arc carries:
A finishing pulse at 3 A for 8 µs carries 0.6 mJ — 330 times less energy, which is the entire reason the surface changes from coarse to fine.
Pulses repeat at a frequency set by the on and off times, and the average electrical power is the pulse energy times that frequency:
A 200 µs on-time with a 100 µs off-time gives a 67% duty cycle. At 40 A and 25 V that is 667 W of spark power. The generator is rated in kVA because reactive and switching losses are real, but the energy that reaches the gap is this number.
The heat does not stay at the surface. For a pulse short enough that heat conduction dominates — which is most of the EDM window — the thermal penetration depth scales as the square root of time:
For H13 tool steel the thermal diffusivity \alpha is about 7 \times 10^{-6} m²/s. At 25 µs, \delta = 13 µm. At 100 µs, 26 µm. At 400 µs, 53 µm. Those three numbers bracket the recast layer thicknesses measured on real parts, which is not a coincidence — the white layer is the part of that thermally disturbed volume that melted and did not get flushed away.
2. The energy budget: 7.6 J/mm³ to melt steel, 2,000 J/mm³ to remove it
Here is the number that reframes the whole process. Melting one cubic millimetre of steel requires heating it from ambient to the melting point and then supplying the latent heat of fusion:
With \rho = 7{,}850 kg/m³, c_p = 460 J/(kg·K), \Delta T \approx 1{,}520 K and L_f = 270 kJ/kg:
Call it 7.6 J/mm³ — and that assumes every melted cubic millimetre is ejected, which it is not. A large fraction of the molten pool re-solidifies on the wall as recast.
Now measure what the process actually draws. A mid-size sinker pulling 10 kW while roughing at 300 mm³/min (5 mm³/s) consumes:
The thermal efficiency is therefore:
Even at the most favourable roughing settings — high current, long on-time, large craters — the figure rarely climbs above 1–2%. The published specific energy consumption for EDM sits at 0.5–1.2 kWh/cm³ on a whole-machine basis, which is 1,800–4,300 J/mm³. Conventional milling of steel needs 2.7–9.3 J/mm³ of cutting energy. The ratio is two to three orders of magnitude, and it is why EDM is only ever justified by geometry, never by throughput.
It also explains a cost fact that surprises first-time buyers: electricity is not the expensive part. In the cavity worked later in this article, the total energy is about 27 kWh — roughly ₹220 at Indian industrial tariffs — against ₹5,250 of machine time. The energy is a rounding error. The hours are the invoice.
3. Electrode wear: the same machine runs at 0.5% and at 70%
The volumetric wear ratio is the electrode volume lost divided by the workpiece volume removed:
This single number drives electrode count, electrode cost and dimensional capability, and it swings by two orders of magnitude across a normal process plan.
At 40 A with a positive electrode (graphite positive, workpiece negative), erosion is dominated by electron bombardment of the workpiece. Graphite loses very little: 0.3–2% by volume, and under 1% on low-wear roughing settings. At 2 A with short pulses, the energy per discharge is small, the plasma spends a larger fraction of its life near the electrode surface, and the same graphite can lose 40–70% of the removed volume. A recent study of fine-grained graphite in the 36–303 µJ discharge-energy range measured a minimum relative wear of 72.9%, and copper at a standard VDI 18 setting measured 70.7%. Finishing wear ratios near 70% are normal. They are not a fault.
Polarity is the control lever. Graphite in positive polarity roughs fast with near-zero wear; the same graphite in negative polarity machines faster but wears at roughly 20% regardless of intensity. In practice: positive for roughing, negative or positive for finishing depending on process stability, and never assume the sign is the same on two different machines.
Stage · Peak current · Gap per side · Undersize per side · Ra after stage · Wear (vol) · MRR
Rough · 20–60 A · 0.15–0.30 mm · 0.25–0.50 mm · 3–6 µm · 0.3–2% · 100–600 mm³/min
Semi-finish · 8–20 A · 0.05–0.10 mm · 0.10–0.20 mm · 1.2–2 µm · 5–15% · 20–60 mm³/min
Finish · 3–8 A · 0.01–0.04 mm · 0.05–0.15 mm · 0.4–0.8 µm · 20–40% · 5–15 mm³/min
Fine / mirror · 1–3 A · 0.005–0.02 mm · 0.02–0.05 mm · 0.15–0.4 µm · 40–70% · 1–4 mm³/min
Planning bands assembled from typical shop technology tables. Your generator, electrode grade and flushing will move these by 30–50%.
Why 30% wear on a finishing pass is catastrophic and 0.5% on a roughing pass is free
The wear ratio is volumetric, so it converts to axial electrode loss by dividing by the same frontal area. If the electrode face area is A and it advances a distance x, the workpiece volume removed is approximately A x and the electrode shortens by \vartheta x:
Roughing: \vartheta = 0.5\% over a 0.40 mm pass shortens the electrode by 2 µm. Irrelevant.
Finishing: \vartheta = 30\% over a 0.05 mm pass shortens it by 15 µm. That is more than the entire ±0.01 mm tolerance band on the cavity. The electrode is out of tolerance before the pass is finished, and no amount of "just run it again" recovers it, because running it again wears it further. This is why the final pass uses a fresh electrode and why the electrode count on a tight cavity is set by the finishing sequence, not by the roughing volume.
4. Undersize arithmetic: where the 0.05 mm corner comes from
The electrode is always smaller than the cavity it burns, by the spark gap plus the orbit excursion plus any stock deliberately left for later passes:
where d_c is the target cavity dimension, s the stock left for subsequent passes, g the spark gap per side and o the lateral orbit excursion per side. For a 100.00 mm cavity with 0.15 mm of stock, a 0.25 mm roughing gap and a 0.10 mm orbit:
Two traps live in this equation.
Double-counting the orbit. Orbiting is not a separate allowance stacked on top of a gap that already includes it. The machine's controller defines the orbit value as a radius, a diameter, or an axial contribution depending on the builder and the orbit shape — circular, square, spherical, vector. Read the definition before you add numbers, and check the swept envelope, not the electrode drawing.
Treating undersize as scaling. An electrode is offset, not shrunk. Every face moves inward by the same per-side amount, which means thin features do not shrink proportionally and a 0.3 mm rib loses 0.6 mm of its 0.3 mm width if you apply a 0.3 mm offset carelessly. Offset the solid; do not scale it.
The corner is the same arithmetic in two dimensions. A sharp internal corner in the cavity requires a sharp external corner on the electrode, and the burn rounds it by the gap plus the orbit:
With a perfectly sharp electrode (r_e = 0), a 0.02 mm finishing gap and a 0.03 mm orbit, the floor is R0.05 mm. In production, R0.1 mm is routine and R0.05 mm is the best case with a fresh fine-grain electrode and no circular orbit. Specify R0.02 mm and no process on earth will hold it, because the spark gap alone is 20 µm.
Corner radius and graphite grade are coupled. A 10 µm average-particle graphite cannot reliably reproduce a 20 µm corner; the grain pulls out. If the drawing needs sharp detail, the electrode material has to be ultrafine or angstrofine — under 5 µm, ideally under 1 µm.
5. Recast: the 30 µm the drawing never mentions
Every EDM surface carries a resolidified layer. It is called the white layer because it resists etching and appears white under an optical microscope. It is fine-grained, heavily alloyed with carbon cracked from the dielectric, harder than the base metal, and in tensile residual stress — the worst possible combination for fatigue.
The thickness tracks pulse energy. Measured on austenitic steel, rough EDM produced an average white layer of 32.09 µm and finishing EDM 9.42 µm — a factor of 3.3, closely following the ratio of the pulse energies. A widely used planning rule is that the white layer is roughly five times the Ra value, and the affected depth beneath it extends further still.
That is why "EDM to Ra 0.4 and leave it" is a complete specification for a mold cavity and an incomplete one for a fatigue-critical part. For anything cyclically loaded, sealing, or subject to impact, the recast has to come off — by lapping, polishing, or a finishing operation that removes more than the layer is thick. The classic guidance is to remove about 1.5× the measured white layer depth to clear the cracks that propagate beneath it. On a 32 µm layer, that is a 50 µm removal — which changes the cavity dimension and therefore has to be planned into the undersize, not discovered afterwards.
Cemented carbide is worse. EDM of WC-Co leaves a cobalt-depleted surface layer, thermal grain cracking, and tensile stress in the recast; the binder can also dissolve when carbide sits in water-based dielectric. For a carbide die that will see sliding wear, the EDM surface is the wear surface, and its integrity matters as much as its geometry. Copper-tungsten electrodes are the standard choice for carbide for exactly this reason: lower wear, more stable burn, less thermal shock at the edge.
6. VDI 3400: the only surface callout that survives a spark
Spark-eroded surfaces are not directional. They are fields of overlapping craters, and Ra alone does not describe them well: two surfaces at Ra 3.2 µm can look different because one has many shallow craters and the other fewer deep ones. The German mold standard VDI 3400 exists to fix that, and it is the callout that toolmakers actually work to.
The scale is logarithmic, which is why every six grades doubles the roughness:
and inverting it:
The tabulated values are rounded — VDI 45 computes to 17.8 µm but is published as 18.0 — so use the formula for interpolation and the chart for callouts.
VDI grade · Ra (µm) · Ra (µin) · Typical character
12 · 0.40 · 16 · Fine EDM surface
15 · 0.56 · 22 · Fine spark texture
18 · 0.80 · 32 · Light matte
21 · 1.12 · 44 · Light cosmetic
24 · 1.60 · 63 · Medium-fine
27 · 2.24 · 88 · Medium
30 · 3.15 · 124 · Medium-coarse
33 · 4.50 · 177 · Coarse
36 · 6.30 · 248 · Coarse matte
39 · 9.00 · 354 · Heavy
42 · 12.50 · 492 · Very heavy
45 · 18.00 · 709 · Extremely coarse
Two practical notes. First, VDI describes roughness, not polish; SPI describes polish. They correlate only approximately through Ra, and a VDI callout does not guarantee a match to an approved plastic part — for cosmetic surfaces, the molded sample is the real specification. Second, a VDI number is a texture family plus a nominal Ra, so specify both the grade and the measurement conditions (instrument, cutoff, evaluation length, direction). A Ra number quoted without those is not a specification.
7. Flushing is the real spindle
The gap is 10 to 300 µm wide, and the debris it must clear is 1 to 20 µm particles of resolidified metal plus carbon. If the local debris concentration rises, the dielectric cannot deionise between pulses, the spark degenerates into a sustained arc, and the arc leaves a pit instead of a crater. That is the failure mode behind most "the burn stalled" and "there are black marks in the corner" complaints.
There are three flushing arrangements and they are not interchangeable:
- Through-spindle / through-electrode — dielectric pushed down the electrode and out through holes, the most effective and the one that requires electrode design to include flushing holes. Bladder or rotating-spindle heads keep the flow axial.
- Suction — dielectric pulled through the electrode, useful for blind pockets where pushing debris deeper is counterproductive.
- Side jet — an articulated nozzle aimed at the gap, the fallback for large cavities and simple electrodes. A controlled study on WC-Co used a side jet at 30 kPa; production pressures are usually set by what the gap will accept without deflecting the electrode or blasting the workpiece.
The geometric consequence is the aspect ratio limit. Die-sinking can hold ribs and slots 0.5–1.5 mm wide at depths many times their width, but only with slow burning, frequent electrode lifts, and often several electrodes per rib. Beyond about 8:1 depth-to-width, flushing access — not the generator — becomes the constraint. Deeper than roughly 15–20:1, the process stops being a burn and becomes a series of controlled stalls.
Electrode lift is part of the cutting strategy, not a housekeeping step. Each retract lets fresh dielectric into the gap and ejects the debris cloud. Short lifts at high frequency for deep ribs; long lifts for large areas with poor access.
8. Machine accuracy and what ISO 11090 actually tests
Sinker EDM tolerances are frequently quoted as ±0.01 to ±0.02 mm on depth and width, with ±0.005 mm achievable using a sizing electrode and a re-measure step. Those numbers are only meaningful if the machine itself is verified, which is what ISO 11090-1:2014 and ISO 11090-2:2014 do: geometric and machining tests, plus accuracy and repeatability of numerically controlled positioning axes, referenced to ISO 230-1 and ISO 230-2. Part 1 covers single-column machines (cross-slide and fixed table); Part 2 covers double-column slide-head machines.
Three things sit outside that standard and routinely blow the tolerance:
- Thermal growth. The dielectric heats up over a long burn. A cavity burned over eight hours is not at the temperature it was at hour one, and the electrode is not either.
- Electrode datum consistency. Multi-cavity tools require every electrode to share one datum. If electrode three is picked up off a different face, the cavity moves.
- Wear during the final pass. Covered above: 30% wear on 50 µm is 15 µm, which is most of the band.
The measurement discipline follows: measure the electrode before the burn, measure the cavity after each stage rather than only at the end, and record the machine's actual spark gap for the settings used. The chart in the manual is a starting point; the gap your machine produces on your dielectric at your temperature is the number that sizes the next electrode.
9. When to EDM, and when it is malpractice
EDM is a capability purchase, not a productivity purchase. The decision is geometric, not economic.
flowchart TD
A[Feature to machine] --> B{Electrically conductive?}
B -- No --> C[EDM impossible]
B -- Yes --> D{Can a cutter reach it?}
D -- Yes --> E{Harder than 45 HRC?}
E -- No --> F[Mill it]
E -- Yes --> G{Deep rib, sharp corner or thin wall?}
D -- No --> G
G -- Yes --> H[Sinker EDM]
G -- No --> I[Hard mill, then EDM only the detail]
H --> J{Depth to width over 8 to 1?}
J -- Yes --> K[Plan flushing and electrode count before quoting]
J -- No --> L[Standard three-electrode burn]
classDef win fill:#e8f5e9,stroke:#1b5e20,color:#111827;
classDef no fill:#fdecea,stroke:#b71c1c,color:#111827;
classDef plan fill:#fff8e1,stroke:#8d6e63,color:#111827;
class H,L win;
class C,F no;
class K plan;
EDM wins when at least one of these is true:
- Hardness above roughly 45–50 HRC, where carbide and CBN cutters start to struggle and cost per cubic centimetre climbs steeply. At 60 HRC, EDM and milling remove material at about the same rate; EDM does not slow down as hardness rises, which is the whole point.
- Sharp internal corners below the smallest practical cutter radius. A 0.2 mm internal fillet in hardened steel has no cutter that can hold it; an electrode reproduces it plus the gap.
- Deep narrow ribs and slots past about 8:1 depth-to-width, where an end mill chatters or breaks.
- Thin walls and fragile features where cutting force would deflect the part. EDM exerts essentially zero mechanical force — the electrode can burn within 0.1 mm of a finished wall.
- Blind cavities and 3D forms that wire EDM cannot reach because there is no through-path for the wire.
EDM is malpractice when:
- You are removing bulk from a soft billet. Milling is 10–100× cheaper per cubic centimetre. Rough before heat treatment, leave 0.2–0.6 mm, then EDM only what the cutter cannot finish.
- The feature is a straight through-cut. Wire EDM is faster, cheaper and more accurate for that.
- The material is non-conductive. No current path, no spark. Ceramics need a conductive assist layer or a different process entirely.
- The tolerance is tighter than the process can hold and nobody has said so. R0.02 mm corners, ±0.002 mm on a deep cavity, and a 0.3 mm rib 20 mm deep are all requests that should come back as questions, not as a quote.
Hard milling has taken real ground — modern carbide grades cut 50–60 HRC — but it has not displaced sinker EDM for deep ribs, sharp internal corners, textured surfaces, and features past 8:1. The economically correct workflow is usually hybrid: mill soft, harden, hard-mill the open geometry, and burn only the details.
10. The cost model: ₹13,750 for one cavity
Take the cavity from the opening: 100 × 80 × 20 mm in H13 at 50 HRC, pre-milled with 0.62 mm of stock. The burnable area is the floor plus four walls, about 15,200 mm². Stock removal by stage:
- Rough: 0.40 mm × 15,200 mm² = 6,080 mm³ at 300 mm³/min → 20 min
- Semi-finish: 0.15 mm × 15,200 mm² = 2,280 mm³ at 30 mm³/min → 76 min
- Finish: 0.05 mm × 15,200 mm² = 760 mm³ at 6 mm³/min → 127 min
- Fine: 0.02 mm × 15,200 mm² = 304 mm³ at 2 mm³/min → 152 min
Total erosion time: 375 minutes, or 6.25 hours. Add 2.5 hours for setup, electrode changes, pickup and measurement — 8.75 machine hours. The last two passes remove 1,064 mm³, which is 11% of the volume and 74% of the spark time. That is the shape of every EDM quote, and it is why a "just skim it finer" request can double a price.
Energy, at realistic average draws of 10 kW roughing, 6 kW semi-finish, 4 kW finish and 3 kW fine:
At ₹8/kWh that is ₹216.
Cost, at an Indian job-shop rate of ₹600/hour for a sinker:
- Machine time: 8.75 h × ₹600 = ₹5,250
- Electrodes: graphite rough ₹1,500, graphite semi ₹1,200, copper finish ₹1,400, copper fine ₹1,400 = ₹5,500
- Electrode milling and programming: 6 h × ₹500/h = ₹3,000
- Total ≈ ₹13,750 for one cavity
Electrodes and electrode machining are 62% of the cost; machine time is 38%; electricity is 1.6%. Two cavities on one electrode set amortise the setup and programming and drop the per-cavity price toward ₹10,500. This is why cavity count matters far more than cavity size in an EDM quote, and why a single prototype cavity in hardened steel is one of the most expensive objects a designer can ask for.
US shop rates for the same work run 80–120/hour, with electrodes at 50–500 each depending on size and grade — the ratios hold, the absolute numbers do not.
11. What to put on the drawing
Most EDM disputes are specification disputes, not machining disputes. A drawing that carries these eight things will quote accurately and burn predictably:
- Material and hardness band after heat treatment — "H13, 48–52 HRC", not "hardened".
- Finish per face, in VDI where texture matters and Ra elsewhere. Mark faces that will be polished, because a polished face has to be burned finer than its final Ra.
- Corner radii as maximums where the corner is functional, and permission to enlarge elsewhere. "R0.05 max" and "R0.3 max" are different jobs at different prices.
- Recast removal on working faces of fatigue, impact or sealing surfaces, with the depth to be removed stated.
- Datums shared across all cavities, and whether electrodes must share one datum.
- Undersize responsibility — whether the toolmaker sizes electrodes from the 3D model or the customer supplies them.
- Flushing access — through-holes, open faces, or a note that the feature is blind and deep.
- The 3D cavity model, not just the 2D drawing. Sinker EDM is a 3D transfer process; a drawing without the solid is a guess.
On FabFlow, jobs that include these details attract quotes in hours instead of days, because the manufacturer does not have to price three assumptions and hope.
12. The short version
- A spark reaches 8,000–20,000 °C for microseconds. Energy per pulse is E_d \approx V_d I_p t_{on}, and thermal penetration scales as \sqrt{\alpha t_{on}} — 13 to 53 µm across the normal pulse window.
- Melting steel needs about 7.6 J/mm³. EDM consumes 500–2,000 J/mm³. Thermal efficiency is under 2%, and electricity is never the cost driver.
- Electrode wear runs from 0.3% at roughing currents to 70% at finishing currents. The same machine, the same night.
- 30% wear on a 50 µm pass shortens the electrode by 15 µm — more than a ±0.01 mm tolerance. The final pass uses a fresh electrode, always.
- Undersize is d_e = d_c - 2(s+g+o) and the corner floor is r_{min} = r_e + g + o. With a 20 µm gap and a 30 µm orbit, R0.05 mm is the physical floor.
- Recast is 32 µm after roughing and 9.4 µm after finishing. On fatigue or sealing surfaces it must be removed, and the removal depth must be planned into the burn.
- VDI 3400 is logarithmic: Ra = 0.1 \times 10^{VDI/20}, +6 grades doubles roughness. Specify grade, Ra, and measurement conditions together.
- Sinker EDM is for hardness, sharp corners, deep ribs, thin walls and blind 3D forms. It is not for bulk removal, through-cuts, or non-conductive materials.
- On a 100 × 80 × 20 mm H13 cavity, expect about 8.75 machine hours and ₹13,750, with 62% of that in electrodes.
The process has not changed much in forty years, and it does not need to. It is the only commercially practical way to put a sharp-cornered, deep, blind shape into steel that is already at 60 HRC — and it will remain the last operation on the routing, the most expensive hour on the quote, and the one that decides whether the tool fits.