Critical engine and transmission torque with AI — a cylinder head tightened out of sequence is a warranty failure.

The problem

Correct torque is not the same as a correct tightening.

On critical engine fastenings (cylinder head, connecting-rod bolts, plastic-elastic fasteners) and transmission ones (gearbox, differential) three conditions have to hold at the same time for the joint to be valid under warranty:

1. Torque inside the window. The nutrunner controller measures it well and, almost always, delivers it well.
2. Correct residual angle. In plastic-elastic tightening, torque is only the start — the angle beyond the threshold torque is what proves the thread went exactly as far as it should.
3. Correct sequence. A cylinder head is tightened in a cross pattern, in increasing passes. Doing it in another order leaves residual stresses that break the gasket at 30,000 km — at the customer, not in the plant.

The classic problem: systems that watch torque do not watch angle and sequence with the same rigor, or they do but the data is never cross-referenced with the specific part or with the VIN. A bolt skipped in the cross pattern, a damaged thread in the block that returns a beautiful torque with a strange angle, a single-use bolt re-tightened by mistake during a touch-up — these are the cases that slip through to the customer and come back as a warranty campaign. That is the most expensive invoice there is in engine manufacturing.

How it fits the IRIS system

iLEAN does not change the tool — it puts torque, angle and sequence in the same file, engine by engine.

The problem is not a missing data point — it is that the three data points live on islands: torque in the nutrunner controller, angle in the same log but read by nobody, sequence in the operator's head and on the process sheet. iLEAN acts as the putty that fills those gaps, without asking you to change the nutrunner or the MES.

Edge sees which bolt is being tightened. Connect reads torque and angle from the nutrunner. The agent cross-references them with the model's sequence spec and, if something does not add up, holds the engine. The person signs — never the other way round.

The three iLEAN pieces applied to critical engine and transmission torque:

• Edge + Vision — a terminal with machine vision (CNN) over the station. It identifies which bolt is being run at each instant (1, 2, 3… in the cylinder head cross pattern) and leaves a timestamped record. It works with no network: if the plant loses WiFi, Edge keeps identifying and, if the sequence breaks, it triggers the actuator.
• Connect — captures torque + angle + time + program code from Atlas Copco, Bosch Rexroth, Cleco or Desoutter controllers over OPC UA, Profinet or EtherNet/IP, at second zero. And it also captures whatever arrives from outside (a recipe change for a fastener batch, an alert from the bolt supplier).
• Agents — cross-reference the triple signature (torque, angle, sequence) with the model's spec, compare it with the engine's VIN/serial and, if something falls out (wrong sequence, short angle, re-tightened bolt), hold the engine before it advances. The person validates and signs; the line never restarts on its own.

See the full IRIS architecture →

Before and after

Classic critical tightening vs. tightening validated by triple signature with iLEAN

Aspect	Nutrunner controller + checklist	With iLEAN Edge + Vision + Connect + Agents
Torque	Measured and stored in a log	Measured + cross-referenced with the recipe + tied to a specific bolt
Angle in plastic-elastic joints	Stored but rarely cross-referenced	Cross-referenced with the model's expected torque-angle curve
Sequence (cylinder head cross pattern)	Operator + paper checklist	Vision identifies bolt by bolt, agent validates the spec
Single-use fasteners	Risk of an undetected re-tightening	Agent blocks a second valid tightening on the same serial
Error detection	Later audit or warranty at the customer	Before it moves to the next station, in seconds
File for 8D / IATF	Rebuilt by hand, weeks	Automatic dossier per engine, with torque + angle + sequence

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 immersion diagnostic.

• Engine or gearbox plant with plastic-elastic tightening on the cylinder head + single-use fasteners on rod bolts; between 15 and 40 critical joints per engine.
• Edge + Vision pilot on one key station (cylinder head or bedplate). First value expected within a few weeks: torque + angle + sequence cross-referenced and the engine held if the sequence breaks.
• Indicative payback between 4 and 9 months, depending on the average cost of a torque-related 8D, the volume of warranty campaigns with "tightening" as root cause and your exposure to a mass campaign.
• Hard lever: reduction of ≥ 30% in incidents with "critical tightening" as root cause (a conservative estimate to be validated). A single engine campaign avoided pays for the system many times over.

The standard that measures whoever audits you

Automotive quality operates in the order of 25 PPM (parts per million) of admissible defects ^[1]. With 15-40 critical joints per engine and volumes of hundreds of thousands a year, the only honest way to defend a low PPM on critical fastenings is to record the triple signature (torque + angle + sequence) per bolt and per engine. The CAIO's reasonable doubt ("what if vision misidentifies a bolt in the cross pattern?") is defused: the AI here is not inventing anything — it is recontextualizing a repeated visual pattern. In anchored tasks, the best models bring error below 1.5% ^[2]. And even so, what is critical is never decided alone: iLEAN holds, the person signs.

^[1] 25 PPM automotive quality standard — Symestic.
^[2] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.