CNC Machining: The Complete Engineering Guide to Milling, Turning, Cutting Physics, Tooling, and Machine Kinematics

A dense equation-driven engineering deep-dive into CNC machining — machine anatomy (cast iron vs polymer concrete, C3 ballscrews, BT40/HSK spindles, servo encoders), cutting physics with worked examples (cutting speed, chip load, MRR, spindle power, Taylor tool life, radial chip thinning, cusp height), the carbide tooling catalog (grades, geometry, coatings, holders, runout), the complete operation library from facing to thread milling, workholding, machinability tables for 11 materials, tolerance capability, the 3/4/5-axis decision, and Indian machining economics with a worked ₹/part quote.

·

CNC Machining: The Complete Engineering Guide to Milling, Turning, Cutting Physics, Tooling, and Machine Kinematics

CNC machining is the reference process of manufacturing. Every other subtractive technology — waterjet, EDM, laser, sheet metal — is measured against it, and every additive technology competes with it. When a bracket must hold ±0.02 mm, a bearing bore must be round within 5 µm, or a prototype must be the same material as production, the part goes on a mill or a lathe.

Yet most engineers interact with CNC through a CAM dialog box and a drawing tolerance, without knowing what happens between the spindle and the chip. This guide closes that gap: the machine's anatomy and why it costs what it costs, the physics of cutting with worked numbers, the tooling catalog that determines what geometry is even possible, the operations library, workholding, materials, tolerances, the axis-count decision, and what a machined part should cost in India in 2026. We have covered the design side in Design for CNC Machining and the programming language in G-Code; this is the process-side companion.

1. A 75-Year Lineage: From Parsons' Templates to 5-Axis

CNC began as an aerospace precision problem. In 1949, John T. Parsons needed to machine helicopter rotor-blade templates with profiles defined by hundreds of coordinates, and the manual method — a machinist reading a table and cranking two handwheels — was too slow and too error-prone. Parsons, working with the MIT Servomechanisms Laboratory, retrofitted a Cincinnati Hydrotel milling machine with servomotors, and in 1952 demonstrated the first numerically controlled (NC) machine, driven by punched paper tape. MIT's APT language followed in 1958, and by the 1970s microprocessors replaced tape readers entirely — NC became CNC.

The physics that made this possible has not changed. What changed is the bandwidth of the feedback loop. A modern 5-axis machine closes a position loop at 250 µs, interpolates hundreds of blocks of lookahead, and holds ±0.005 mm with glass-scale feedback at 0.1 µm resolution. A machine tool is, at bottom, a stiffness problem wrapped around a servo problem wrapped around a cutting problem. Keep that ordering in mind — it explains everything downstream.

2. Machine Anatomy: Five Subsystems, Five Budgets

Every machining center, from a ₹6 lakh used VMC to a ₹4 crore 5-axis, is five subsystems. Understanding them is how you read a machine spec sheet and how you diagnose why a part is out of tolerance.

flowchart TD
    A[CAD model] --> B[CAM: toolpaths, feeds, speeds]
    B --> C[Post processor: G-code]
    C --> D[CNC controller: interpolation, lookahead]
    D --> E[Servo drives: position & velocity loops]
    E --> F[Ball screws / linear motors: motion]
    F --> G[Spindle + tool: chip formation]
    G --> H[Inspection: CMM / in-machine probing]
    H -->|feedback| A
    style A fill:#ffffff,stroke:#000000
    style G fill:#1e3a5f,stroke:#38bdf8,color:#ffffff

2.1 The Frame: Stiffness and Damping

The machine's structure must resist cutting forces without deflecting more than a few micrometres, and it must damp vibration rather than ring. Cast iron (typically FC25–FC30 grey iron) is the standard: good stiffness (E ≈ 110–120 GPa), good damping, and — critically — thermal mass. Premium machines use polymer concrete or epoxy granite (e.g., mineral cast bases): modulus is only ~40 GPa, but damping is 8–10× better than cast iron, which is why it kills chatter in high-speed finishing.

The design rule: static stiffness of a production VMC frame is on the order of 50–100 N/µm. At a 500 N cutting force, 100 N/µm gives 5 µm of deflection — right at the edge of a ±0.01 mm tolerance. This is why "machine rigidity" is not marketing.

2.2 Linear Motion: Ballscrews and Guides

The workhorse actuator is the preloaded ballscrew. ISO 3408 grades them by travel accuracy per 300 mm:

Grade · Deviation per 300 mm · Typical use

C1 · ~6 µm · Jig borers, ultra-precision

C3 · ~8 µm · Precision machines, mold work

C5 · ~18 µm · Standard production VMC/lathe

C7 · ~52 µm · Routers, woodworking, budget CNC

Linear guides (profiled rails with recirculating ball carriages) come in P/H/N accuracy classes; P-class rails hold ~±0.005 mm/m straightness. High-end machines dispense with screws entirely and use linear motors (direct drive): no backlash, no wear, 2 g acceleration — at 2–4× the cost.

2.3 The Spindle

The spindle is a motorized precision bearing assembly rated by taper, speed, and power. Tapers:

Spindle bearings are the precision heart: angular-contact sets in a P4/ABEC-7 class, preloaded, oil-air lubricated at high speed. Bearing DN limits (bore mm × rpm) cap speeds: a 70 mm bore at 15,000 rpm is DN = 1.05M — already into hybrid ceramic territory. Ceramic balls (Si₃N₄) extend DN to 1.5M+.

2.4 Drives and Control

A modern AC servo closes three loops: current (µs), velocity, and position (100–250 µs). Encoders are 20–24 bit — 2²⁰ = 1,048,576 counts/revolution, i.e. 0.00034° resolution. On a 10 mm pitch screw, one count is 9.5 nm. The loop bandwidth and lookahead (200–1,000 blocks) determine whether the machine can hold accuracy at speed — a machine that's accurate at 500 mm/min can be rubbish at 5,000 mm/min.

Controllers: Fanuc (0i/30i series), Siemens SINUMERIK (828D/One), Heidenhain (TNC series, preferred for 5-axis mold work), Haas NGC, Mitsubishi M80, and budget Syntec/LNC — the last is common on Chinese and Indian-built machines.

2.5 Thermal Reality

Steel grows ~12 µm/m/°C, cast iron ~10, aluminum ~23. A 600 mm steel column with a 4°C temperature gradient across the day moves 29 µm — two-thirds of a precision tolerance, from doing nothing. This is why serious machines have spindle chillers, thermally symmetric frames, and warm-up cycles (15–20 min at low speed before fine work), and why CMM rooms hold 20 ± 1 °C. When a machinist says "let the machine warm up first," they're doing metrology, not ritual.

3. The Three Kinematic Families

3-axis VMC (vertical machining center). Tool moves in X/Y/Z; the part sits in a vise. Two visible faces per setup; each additional face needs a new setup with its own alignment error. 90% of prismatic parts live here.

Turning (lathe). The part spins (up to 4,000+ rpm on bar stock), the tool moves in X/Z. Chucks, collets, bar feeders, subspindles, live tooling (C-axis milling on the same machine = "mill-turn"). Any revolved feature — shafts, bushes, pulleys, fittings — is a lathe part. Turning is cheaper than milling because metal removal is continuous, not interrupted.

4- and 5-axis. The 4th axis is usually a rotary table (A) on a VMC for indexing parts to new faces without refixturing. True 5-axis (A/B + C trunnion or tilting head) adds simultaneous motion, which enables two things 3-axis can never do: machining complex sculpted surfaces with a ball-nose always normal to the surface, and cutting with short tools. On a 5-axis machine you can tilt the part and reach a deep pocket with an L/D = 3:1 tool where a 3-axis would need 6:1 — and tool deflection scales with the cube of length-to-diameter, so the 5-axis tool is 8× stiffer.

4. Cutting Physics: The Four Equations Every Machinist Lives By

Everything in a CAM speed/feed dialog reduces to four quantities. Units: D = tool diameter (mm), N = spindle speed (rpm), z = number of flutes, f_z = chip load (mm/tooth), v_f = feed rate (mm/min), a_p = axial depth of cut, a_e = radial engagement (mm).

Cutting speed — the surface speed of the tool edge, set by material/tool pairing:

Feed rate — how fast the tool advances, set by chip load:

Material removal rate — the money equation; MRR is what a job shop sells:

Spindle power — from MRR and the material's specific cutting force k_c (N/mm²):

and torque, which limits low-rpm roughing more than power does:

Worked example. A 12 mm 2-flute carbide end mill in 6061-T6: v_c = 500 m/min → N = 500,000/(π × 12) ≈ 13,263 rpm. With f_z = 0.06 mm/tooth: v_f = 0.06 × 2 × 13,263 ≈ 1,592 mm/min. At a_p = 6 mm, a_e = 2.4 mm (20% of D): Q = 6 × 2.4 × 1,592/1,000 ≈ 22.9 cm³/min. Power at k_c ≈ 800 N/mm²: P = 22,925 × 800/(60 × 10⁶) ≈ 0.31 kW. An 11 kW spindle is not there for roughing aluminum — it's there for torque at low rpm in steel.

Taylor's tool-life law is the one equation that disciplines speed selection. Tool life T (min) and cutting speed relate as:

with n ≈ 0.125 for HSS and n ≈ 0.20–0.30 for carbide in steel. Rearranging: T_2/T_1 = (v_1/v_2)^{1/n}. With n = 0.25, doubling the cutting speed cuts tool life 16×. A 20% speed increase halves it. This is why machinists treat aggressive speed/feed tables with suspicion — the tooling rep's chart optimizes tool cost, but the shop optimizes total cost per part.

Radial chip thinning. When a_e < D/2 (the normal case in finishing and HSM), the programmed chip load f_z is not what the tooth actually cuts. The true maximum chip thickness is:

At a_e/D = 0.2, φ = arccos(0.6) ≈ 53.1°, so h_m = 0.8·f_z. To keep the real chip load constant, feed must be raised as engagement drops — this is why trochoidal paths run 2–3× the nominal feed. Failing to compensate means the tooth rubs instead of cuts: rubbing work-hardens the surface, and in stainless that's a death spiral of heat, built-up edge, and broken tools.

Surface finish from geometry. A face mill or ball-nose leaves scallops. For a corner radius r_ε (mm) and feed f_z (mm/tooth), the theoretical finish is:

A 12 mm end mill with r_ε = 0.8 mm at f_z = 0.04 mm: Ra ≈ 0.04²/(6.4) = 0.25 µm — but only if the machine is rigid and the tool is sharp; real shop finishes land at 2–4× theoretical. For 3D surfaces, ball-nose stepover s leaves a cusp of height h = s^2/(8R): with R = 4 mm and s = 0.5 mm, h ≈ 7.8 µm. Halving the stepover quadruples the finishing time — surface finish is bought with quadratic time.

5. Tooling: The Catalog That Defines Your Part

5.1 Carbide Grades

Solid end mills are tungsten carbide: WC grains (~0.2–0.8 µm, submicron grades for sharp edges) in a cobalt binder (6–12%). Hardness 90–93 HRA versus 63–66 HRC for HSS; stiffness ~550 GPa versus ~210 GPa. Fine grain = sharp edge for aluminum; coarse/ultra-tough grades take interrupted cuts in steel. Beyond carbide:

5.2 Coating Selection

Coating · Color · Max temp · Best for

Uncoated polished carbide · silver · — · Aluminum, non-ferrous (no built-up edge)

TiN · gold · ~600 °C · General steel, first-coating default

TiCN · grey-violet · ~400 °C · Abrasive wear, cast iron

TiAlN / AlTiN · dark violet · ~900 °C · Dry high-speed steel, stainless, hardened

AlCrN · blue-grey · ~1,100 °C · Extreme heat, Ti and Ni alloys

DLC · black · ~400 °C · Anti-galling on aluminum, thin

Rule: aluminum gets uncoated polished carbide (or DLC); steel gets TiAlN for dry high-speed or TiN for wet general work. Coatings on aluminum tools cause galling and built-up edge — a coated tool can perform worse than uncoated in non-ferrous.

5.3 Geometry: Flutes, Helix, Corner

5.4 Tool Holders and Runout

Holder · Runout TIR · Notes

Sidelock (Weldon) · 10–20 µm · Cheap, brutal, fine for roughing

ER collet · 5–10 µm · The default; good collets, tightened properly, hit ~5 µm

Hydraulic · 2–3 µm · Oil sleeve grips the shank; great price/performance

Shrink-fit · 1–3 µm · Heat the bore, shrink onto the shank; shortest gage length

Milling chuck · 3–8 µm · High gripping torque for roughing

Runout is a tool-life tax. At 10 µm TIR, one flute cuts ~10 µm deeper than its opposite number — the whole chip load lands on half the flutes. Industry data consistently shows ~50% tool-life loss per 10 µm of runout, and surface finish degrades with it. A ₹12,000 hydraulic holder that preserves a ₹3,000 carbide tool's life is not an accessory — it's insurance.

6. The Operation Library

Milling. Facing (face mill or large end mill — the first op on most parts); slotting (full-width a_e = D, the hardest cut: interrupted, chip re-cutting, 360° engagement); pocketing (zig-zag is fastest, offset/spiral gives better finish and wall parallelism; climb milling throughout); contouring (peripheral finishing); HSM/trochoidal (constant low engagement a_e = 5–10% D with deep a_p — heat leaves in the chip, tools last longer, MRR often beats slotting); plunge roughing (Z-axis only; the stiffest direction of the machine, for deep pockets); chamfering/engraving.

Drilling family. Twist drill (118° general / 135° split-point for hard materials) to 5×D normally, 8–10×D with pecking (G83), carbide coolant-through to 20×D+. Reaming for IT7 holes; boring for true position (a bored hole is coaxial with the spindle — the fix for drill wander). Tapping vs thread milling: taps are 3–5× faster, but a broken M3 tap buried in a ₹40,000 part is a disaster; thread milling (single-tooth, circular interpolation) breaks as powder, makes any pitch, and is mandatory in titanium, hardened steel, and deep blind holes. Standard engagement: 1.5×D in steel, 2×D in aluminum, 2.5×D in soft brass/plastic.

Turning. Roughing with CNMG/WNMG inserts at a_p = 2–6 mm; finishing with DCMT/VCGT at 0.2–0.5 mm; constant surface speed (G96) so the insert sees the same v_c as diameter shrinks; parting off with 2–3 mm blades; knurling; single-point threading. Finish on the lathe from geometry: R_a \approx f^2/(8 r_\varepsilon) with f in mm/rev — an r_ε = 0.8 mm insert at f = 0.15 mm/rev gives Ra ≈ 3.5 µm; drop to f = 0.08 → Ra ≈ 1 µm.

Climb vs conventional. Climb milling (cutter rotation feeding the tooth into the cut, chip thick→thin) puts the force into the machine rather than lifting the part, leaves better finish, and is the default on any CNC. Conventional (thin→thick) is for manual machines with backlash and for flame-cut edges with hard scale.

7. Workholding: Where Parts Get Scrapped

The vise looks trivial and decides everything. Standards: a 6" Kurt-style machined vise (repeatability ±0.02 mm with parallels); soft jaws — aluminum jaws machined in-situ to the part's exact contour, the single highest-ROI workholding upgrade (repeatability to ±0.005 mm, no part marking); dovetail prep (a 45° dovetail milled into the stock blank, gripped by matching jaws — the standard op-1 strategy for 5-axis parts so op-2 can reach 5 faces); Mitee-Bite style side clamps and toe clamps for plate work; vacuum plates and double-sided tape for thin (<3 mm) sheet; collets and chucks on the lathe; and zero-point pallet systems (Lang-style, ~5 µm repeatability) that let setups happen offline. The golden rule: the fixture must not require the machinist to measure anything twice, and it must never depend on a single datum you just machined away.

8. Materials and Machinability

Machinability ratings below are relative to free-cutting B1112/1212 steel = 100%, and speed ranges are production carbide values (HSS runs roughly ⅓ of these).

Material · Rel. machinability · v_c carbide (m/min) · f_z rough, 10 mm EM · k_c approx (N/mm²)

6061-T6 aluminum · ~270% · 300–800 · 0.05–0.10 · ~800

7075-T6 aluminum · ~150% · 250–600 · 0.05–0.08 · ~800

Brass C360 · ~100% · 200–400 · 0.05–0.08 · ~700

1018 mild steel · ~75% · 120–220 · 0.04–0.08 · ~1,900

4140 (28–32 HRC) · ~65% · 100–180 · 0.04–0.06 · ~2,200

304 stainless · ~45% · 50–120 · 0.03–0.05 · ~2,400

316 stainless · ~35% · 45–100 · 0.03–0.05 · ~2,600

Ti-6Al-4V · ~25% · 40–80 · 0.03–0.05 · ~1,800

PEEK · — · 100–250 · 0.05–0.08 · low

Acetal (POM) · — · 150–300 · 0.05–0.10 · low

Nylon / HDPE · — · 200–400 · 0.05–0.12 · low

The two materials that punish the unwary: 304/316 stainless (work-hardens instantly — a rubbing pass with no chip turns the surface to armor, and the next pass smokes the tool; feeds must never drop below ~0.03 mm/tooth) and Ti-6Al-4V (thermal conductivity ~7 W/m·K versus ~150 for 6061 — the heat has nowhere to go except into the tool, so speeds stay at 40–80 m/min and MRR stays low; the machinist's compensation is that titanium is machined with HSM strategies at small a_e, where chip evacuation does the cooling). Plastics want razor-sharp tools, positive geometry, and — for PEEK — annealing before and after machining to relieve residual stress, or the part will creep out of tolerance on the bench.

9. Tolerances: What a Machine Can Actually Hold

Shop-class capability, not sales-brochure capability:

Process · Typical tolerance · Tight (skilled shop) · Achievable finish Ra

3-axis milling · ±0.1 mm (ISO 2768-m) · ±0.02 mm · 0.8–3.2 µm

Precision milling · ±0.03 mm · ±0.005 mm · 0.4–1.6 µm

Turning · ±0.05 mm · ±0.01 mm · 0.8–1.6 µm

Boring · IT7 · IT6 (≤ ±0.005 mm on Ø20) · 0.4–0.8 µm

Grinding (follow-on) · ±0.005 mm · ±0.002 mm · 0.1–0.4 µm

Two structural caveats. First, thermal — the numbers above assume a warm, temperature-controlled machine; on a 35 °C Indian shop floor at 4 pm versus 9 am, a 300 mm aluminum part moves ~0.02 mm from its own expansion alone. Second, feature count — a shop that holds ±0.02 mm on a 5-feature bracket will not hold it on a 50-feature manifold at the same price; every tight dimension needs its own tool pass, its own inspection, and its own risk margin. Our GD&T and tolerance stackup guides cover how to specify only the tolerances that matter.

10. 3-Axis vs 4-Axis vs 5-Axis: The Decision

Choose 3-axis when: prismatic parts with features on ≤3 faces, quantities where two setups are acceptable, budget is the constraint. Hourly rates are the lowest, shop capacity is everywhere, and tolerance risk is lowest per setup.

Choose 4-axis (or 3+2) when: the part has features on 4+ faces that must relate to each other. Indexing removes a setup and its stack-up error. 3+2 positioning on a 5-axis machine gives most of the benefit without the programming cost of simultaneous motion.

Choose simultaneous 5-axis when: (a) sculpted/organic surfaces need a ball-nose normal to the surface; (b) deep features need short rigid tools (L/D ≤ 3:1) — impellers, blisks, mold cores; (c) the tolerance budget cannot absorb a second setup — a 5-axis part is machined in one holding, so the entire datum system is established by the machine, not by a re-fixturing operation. The price: machines cost 2–5× a comparable 3-axis, programming is 2–3× harder (a 5-axis collision is a six-figure event), and shop rates run ~2×. Most parts — including most aerospace brackets — are still 3-axis parts.

11. Economics: What a Machined Part Should Cost

A machine-shop quote is a five-term sum:

Worked example. 6061 bracket, 25 pieces. Cycle 8 min, rate ₹1,200/hr → cycle cost ₹160. Setup 2 h at ₹1,200 = ₹2,400 → ₹96/part. Material 0.4 kg at ₹300/kg + 25% waste → ₹150. Tooling ₹10, QC ₹15. Total ₹431; with 25–35% margin the quote lands at ₹540–580/part. Order 250 instead of 25: setup drops to ₹9.6/part and the machinist can afford a fixture and an optimized toolpath — the quote might fall 35–40%. This quantity-vs-setup curve is the entire reason batch size appears on RFQ forms, and why you should never judge a machine-shop quote by the raw cycle time.

Against 3D printing: for aluminum parts, CNC beats FDM/SLS on cost and material properties from Qty ≈ 1 upward — the crossover is about complexity, not quantity. A part with internal lattices is un-machinable; a prismatic bracket is cheaper machined at quantity one. Plastics flip the math: SLS nylon wins at Qty 1–10 for complex geometries, injection molding at 1,000+, with CNC in between for tight tolerances and specific polymers. See our 3D-printing-vs-injection-molding economics for the crossover math.

12. The DFM Essentials (Quick Reference)

The full treatment is in Design for CNC Machining; the five rules that prevent 80% of requotes:

  1. Internal corners need radii — the tool is a cylinder. R ≥ D/6 min (an R3 for an 18 mm feature), R ≥ ⅓ × pocket depth for deep pockets, or the shop quotes a tiny end mill at 4× cycle time.
  2. Depth limits — pocket depth ≤ 3×D for standard tools, ≤ 4×D necked, >10×D = special-order and expensive. Wall height-to-thickness ≤ 8:1 or thin walls sing.
  3. Holes — standard diameters, through where possible; blind holes need extra depth for the drill point (add ~0.3×D); tapped holes ≤ 2×D engagement unless you specify deeper.
  4. Avoid undercuts and dovetails — they need T-slot cutters or a 5-axis op; both cost.
  5. Don't over-tolerance — every digit after the decimal is a line item. ±0.1 mm everywhere and ±0.02 on the two bores is the difference between a ₹500 part and a ₹1,500 part with identical function.

13. CNC in India: Where the Capacity Lives

Indian machining clusters have genuine world-class pockets: Pune (automotive precision, toolrooms), Bengaluru (aerospace, defense, deep 5-axis and EDM capacity), Rajkot (turning and castings at commodity prices), Coimbatore (pump and textile machinery precision), Ahmedabad, Gurgaon, Chennai, Belagavi (general job work). Machine populations run the full spectrum — Haas, DMG MORI, Doosan, Mazak, and Makino imports alongside domestic builders (Ace Micromatic, Jyoti, BFW, LMW, and the newer Lakshmi/Ace 5-axis lines) and large Chinese-import VMC fleets. Skill is the constraint, not hardware: a competent 5-axis programmer in India is rarer than a 5-axis machine.

For buyers, the practical advice is brutal and simple: quote from the drawing, not from the photograph. A clear 2D drawing with GD&T, material, quantity, and finish beats a 200 MB STEP file with no tolerances — the first gets an accurate quote in 24 hours, the second gets a "call us" and a 40% risk premium. On FabFlow, verified CNC shops bid on exactly these parameters — machine capability, material, tolerance, and batch size are structured fields, not free-text, which is why quotes come back comparable instead of scattered.

14. References

Need machined parts? Post your job on FabFlow with your drawing and tolerances — verified CNC shops across Pune, Bengaluru, Rajkot, and Coimbatore quote against it, and the platform tracks the job from quote to shipment.

More FabFlow blog posts