Crankshaft machining with AI — the spindle says it long before the metrology room does.

The problem

The defect is born at the spindle and is discovered at the CMM, one OP60 later.

A crankshaft goes through dozens of operations — crankpin machining, main journal machining, keyway broaching, oil gallery drilling, grinding, dynamic balancing — and the truth about the dimensions is signed off in the room CMM. By the time that CMM says "this part is out of spec", the part already carries machine-hours that will never come back.

1. The tool ages silently — CBN, PCD, ceramic. Expensive, durable and treacherous: it holds and holds, until a micro-edge goes and the surface finish goes with it.
2. The spindle changes its signature — the vibration and the acoustics of a normal operation drift before the dimension does, but that data never gets cross-referenced with the part that was in the chuck.
3. The on-machine probe measures the dimension — but point by point, and at the end of the cycle. If the problem showed up two thirds of the way through, a part is already gone.
4. The operator is running six machines — the veteran's intuition hears the change in the grinder before anyone else, but he cannot be on six shifts at once.

The irony: the signals that predict the drift (vibration, spindle current, acoustics) are right there. The CNC uses them to regulate, not to warn.

How it fits the IRIS system

iLEAN does not replace the CNC or the CMM — it adds the brain that cross-references signal with dimension and with batch.

The classic problem in crankshaft machining is information on islands: the CNC holds the process data, the on-machine probe holds the dimension, the room CMM holds the ground truth, and nobody cross-references the three in real time. iLEAN is the putty that fills the joint between those three worlds.

Edge listens to the spindle. Connect reads the CNC and the tooling ticket. The agent cross-references signal ↔ dimension ↔ history and warns before the next OP. The person decides the change.

The three iLEAN pieces applied to crankshaft machining:

• Edge — an at-machine terminal with an accelerometer and a microphone (where needed), running a neural network trained to recognize the signature of each operation (crankpin grinding, counterweight milling, keyway broaching). When the signature drifts outside the learned band, it raises the alert. It works with no network: if the plant loses WiFi, Edge keeps listening and holds the next part if the drift is confirmed.
• Connect — captures the CNC data (Fanuc, Siemens, Heidenhain, Mazak) through whichever mode applies (modern integration or reading from the machine PC), captures the tooling ticket from the tool crib, captures the room CMM reports. It also captures what arrives from outside: the email from the grinding wheel supplier about a mix change, the WhatsApp from the shift lead noting an odd noise.
• Agent — cross-references the vibration/acoustic signature with the on-machine probe reading, with the CMM history and with tool life. When it sees the pattern starting to look like the known "tool about to break" pattern, it holds the part as soon as it leaves the OP and alerts the operator and the process engineer. It does not change the tool on its own — the person validates and signs.

See the full IRIS architecture →

Before and after

SPC + final CMM control vs. predictive control with iLEAN

Aspect	SPC + on-machine probe + CMM	With iLEAN Edge + Connect + Agent
Moment of detection	End of cycle (probe) or metrology room (CMM)	During the cycle, from the spindle signature
Root cause of the defect	Rebuilt by hand from hypotheses	Automatic hypothesis: "loaded wheel edge at OP30"
Tool change	By theoretical hours + the veteran's intuition	By the real signature, neither sooner nor later
Scrap part discovered at OP60	OP20-OP50 wasted on a part already lost	Held as soon as the pattern looks suspicious at OP20
Operation without network	n/a	Edge keeps running with its own light
File for IATF 16949 / OEM customer	Rebuilt manually by batch	Dossier by crankshaft number with all its signatures

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Crankshaft machining line with OP10-OP60, a mix of CNC grinders, broaching machines and a machining center, with a Zeiss/Hexagon room CMM as referee.
• Edge pilot on 2-3 critical machines (crankpin grinder + keyway broaching + one finishing OP). Accelerometer on the spindle + CNC data reading. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on how often dimensional/finish scrap has occurred in recent months and on the average cost of a part with OP30+ already on it.
• Hard levers: scrap avoided on parts carrying machine-hours + extended CBN/PCD tool life + an automatic dossier by crankshaft number for the OEM.
• The automotive standard of around 25 PPM ^[1] does not hold up with the final CMM as the only filter.

And the machining engineer's reasonable doubt

"What if the AI mistakes a harmless vibration for an edge breaking and stops a healthy machine?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI compares a signature against patterns trained on your own machines, the best models brought error below 1.5% ^[2]. And even so, what is critical is never decided alone: the system warns and holds, the person decides. iLEAN's three safety rings exist precisely for this.

^[1] Symestic — automotive quality standard of ~25 PPM.

^[2] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.

Related processes: critical torque in engine and transmission · stamping dimensional control with 3D vision · predictive maintenance for KUKA/Fanuc robots.