A single stopped station on a final assembly line can cost a large automaker roughly $10,000 to $20,000 per minute in lost output (a widely cited industry estimate). The car that was 40 seconds from rolling off now sits frozen because a $3 wiring harness never arrived at the right station.
Here is the uncomfortable truth: the data that could have predicted that shortage existed hours earlier. It was sitting in a supplier's shipping system, in a plant sensor, in a scheduling database. Nobody connected the dots in time.
This lesson walks a vehicle from raw steel to finished car and shows exactly where data exposes bottlenecks and warns you before the line goes down.
A vehicle is built in a rough sequence: stamping (pressing sheet metal into panels), body shop (welding panels into a frame), paint, then final assembly (installing engine, interior, wiring, glass). Each stage generates a different data exhaust.
Giant presses stamp doors, hoods, and floor pans. Each press reports cycle time (seconds per stroke), tonnage, and die temperature.
When cycle time drifts up by even a fraction of a second per stroke, that is a throughput signal. Multiply a small delay across thousands of strokes per shift and you lose real parts. Rising tonnage variance often means a die is wearing, which predicts a defect or an unplanned stop.
The body shop is mostly robots spot-welding. Here the data comes from IoT sensors (small networked devices measuring vibration, current draw, temperature) attached to robots and weld guns.
A weld gun drawing more current than usual, or a robot arm vibrating outside its normal band, is an early failure signature. Catch it and you schedule a fix during a planned break. Miss it and the robot fails mid-shift and starves every downstream station.
Paint is often the slowest, most expensive, most rigid part of the plant. Ovens must stay hot, booths must stay clean, and color changes waste material and time.
Paint data (booth humidity, oven temperature, flow rates, first-pass yield) matters because rework here is brutal. A car that fails paint inspection goes back through the line, consuming a slot another car needed. First-pass yield (the percentage of cars painted correctly the first time) is a top throughput lever.
Now the body meets thousands of purchased parts: seats, harnesses, dashboards, tires. This is where supply-chain data collides with plant data.
Here is a core operations idea, often associated with the Theory of Constraints: a system's output is set by its slowest step. Speeding up any station that is not the bottleneck just piles up inventory in front of the real constraint.
Data's first job is to *find* the bottleneck, and the constraint moves. Today it is paint. Tomorrow a supplier delay makes final assembly the constraint. You cannot manage what you cannot see in near real time.
A simple MES query exposes it: which station has the longest queue and the highest utilization?
-- Find the current bottleneck station by wait time
SELECT station_id,
AVG(wait_seconds) AS avg_wait,
AVG(utilization_pct) AS avg_util
FROM mes_station_events
WHERE event_time > NOW() - INTERVAL '2 hours'
GROUP BY station_id
ORDER BY avg_wait DESC
LIMIT 5;The station at the top, high wait and high utilization, is where every improvement dollar should go first.
For a deeper grounding in constraint thinking, the Wikipedia overview of the Theory of Constraints is a solid free primer.
Final assembly runs on parts arriving in the right order at the right minute. The data backbone for this is EDI (Electronic Data Interchange): a standardized format automakers and suppliers use to exchange business documents (purchase orders, shipping notices, invoices) machine to machine.
The key document is the ASN (Advance Shipping Notice): an EDI message a supplier sends saying "this shipment, these parts, this quantity, arriving at this time." (In EDI's common standard, this is transaction set 856.)
Predicting a stoppage is a race between two clocks:
When (on-hand + confirmed inboundinboundA strategy that attracts prospects organically via valuable content (blog, SEO, social) rather than interrupting them.Voir la définition complète →) divided by burn rateburn rateBurn rate is the speed at which a company spends its cash reserves, usually measured per month, before reaching profitability or raising more funding.Voir la définition complète → drops below the transit-and-handling time, you have a predicted shortage. That is your alarm, and it can fire hours before the shelf goes empty.
The warning almost never comes from your own warehouse count alone. It comes from *combining* sources:
Any one of these alone looks harmless. Together they predict a 3pm stoppage. This is why connecting plant-floor data (MES, IoT) to supply-chain data (EDI, telematics) is the whole game.
Automotive pioneered just-in-time (JIT): keeping minimal inventory and having parts arrive exactly when needed. JIT slashes cost but removes the cushion. With little buffer stock, a late truck becomes a stopped line fast.
The 2020 to 2023 semiconductor shortage taught the industry this hard lesson: many plants idled because a few dollars of chips were missing. The response has been more data visibility deeper into the supply chain, not just at your tier-one suppliers but the tiers beyond them.
Jargon check: a tier-one supplier sells directly to the automaker (say, a seat maker). A tier-two sells to the tier-one (the foam or fabric maker). Shortages often start two or three tiers down where you have the least visibility.
Vérification des acquis
1. The lesson emphasizes that the data needed to predict a part shortage often exists well before a line stops, but sits unconnected across different systems. What core problem does this illustrate?
2. In stamping, why is a rising cycle time of even a fraction of a second per stroke treated as a meaningful throughput signal?
3. How is the MES (Manufacturing Execution System) best conceptually described based on the lesson?
4. Select ALL correct answers. Which of the following are examples of using data as a predictive (rather than merely reactive) tool, as described in the lesson?
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers. Which statements accurately reflect the role of IoT sensors in the body shop?
Sélectionnez toutes les réponses correctes.
Data only creates throughput when it drives an action. Three concrete plays:
If a specific part is short but you have bodies that do not need it (a base trim without the premium sound system), the MES can resequence the line to build those cars now and delay the affected ones. You keep the line moving instead of stopping it.
A predicted shortage lets you make a cheap decision early (reroute a truck, pull from another plant) instead of an expensive one late (air-freight parts overnight). The value of the prediction is the gap between those two costs.
If paint first-pass yield drops every time humidity crosses a threshold, that pattern justifies a booth control upgrade. If a supplier's parts fail inspection at a rising rate, EDI quality data gives you the evidence to act. The goal is a closed loop: sense, predict, act, measure.
Mature automotive data operations share a few traits:
You do not need a perfect data lakedata lakeA data lake is a centralized repository that stores large volumes of raw data in its native format, from structured tables to unstructured files, until needed.Voir la définition complète → to start. You need the burn rateburn rateBurn rate is the speed at which a company spends its cash reserves, usually measured per month, before reaching profitability or raising more funding.Voir la définition complète →, the on-hand count, and the ASN, connected and compared, for your single most constrained part. That one loop pays for itself the first time it prevents a stoppage.