Skip to content
woodfrog

·

Analytics
Automotive, car-body manufacturing, IndiaCase study draft, anonymised

Catch the bad weld before the body leaves the plant.

The only way the plant could prove a weld was good was to break it. So it broke a few from each batch and trusted the rest. Now every weld is measured as it is made, and an alert names the exact car to check.

Robots welding car bodies on the line
Robots welding car bodies on the assembly line
Every weldMeasured as it is made, not sampled
One carNamed by each alert, not a whole batch
3 signalsVoltage, temperature and cycle time
In the plantAlert raised before the body leaves
WhoA car-body manufacturer, at its plant in India. Name not published.
BuiltData capture from the welding line, each weld tied to its car, patterns from past failures, and an alert per car
AreaAnalytics: command centres and decision applications. Data foundations: data pipelines.
The weld that tells on itself

From a batch you sample to a car you can name.

Every weld now leaves a record as the body passes the station. All the traces run steady, except one. Only that car lights up.

Before and after

From breaking a few welds to reading all of them.

Before
  • A few bodies per batch were broken to test their welds
  • Every weld that was not broken was trusted, not checked
  • A failed sample put a whole batch under suspicion
  • The machines saw every weld, and none of it was kept
After
  • Every weld is recorded as it is made
  • Each reading belongs to a part and a car
  • Known warning signs are spotted automatically
  • An alert names one car to check, while it is still in the plant
The problem

The only proof of a good weld was to break it.

On a car-body line, a weak weld can look exactly like a good one. The plant took bodies from a batch at random and broke their welds to see if any gave way. A failed sample put the whole batch under suspicion, without saying which car was at fault.

Robotic arm welding, sparks flying
Every weld on the line, and none of it was being kept
What was broken
  • Testing a weld meant destroying it, so most welds were never tested.
  • A failed sample raised doubt about a batch, not one car.
  • The welding machines saw voltage, temperature and cycle time for every weld. None of it was kept.
  • Nobody could say what was different about the welds that failed.
You might be here if
  • Your quality check destroys the part it checks.
  • One failed sample puts a whole batch on hold.
  • Your machines see every cycle, and none of that data is kept.
  • Nobody can say what was different about the parts that failed.

Why it mattered. A body that leaves the plant carries its welds into a finished car. With sampling, the plant could say a batch looked fine. It could not say every car was.

The approach

Stop sampling the welds. Listen to all of them.

The machines already knew the conditions of every weld. The work was to keep that data, tie each reading to its part and car, and compare it with the welds that had failed before. Where failures lined up with the voltage dropping or the cycle running short, that pattern became the trigger for an alert.

Robotic arms on a car assembly line
The welding line, already in place
The line, already in place
Welding machinesEvery weld, every station
Part sequenceWhich part, which car
Weld issue recordsFrom the destructive tests
Built from scratch by woodfrog
Capture every weldVoltage, temperature, cycle time
Tie each weld to its car
Patterns from past failuresVoltage dropping, cycle running short
Instant alertCheck the welds on this car
On the line
Quality teamChecks the named car
One car named, not a batch
The old answerA whole batch under suspicion
Past weld issues teach the system what a bad weld looks like. The live path runs from the machines to the alert.schematic
The solution

Built from scratch, starting at the machines.

What was built
  • Data capture from the welding machines, from scratch: voltage, temperature and cycle time for each weld.
  • Every weld tied to the part sequence, so each reading belongs to a part and a car body.
  • Weld data matched to known welding issues, to find what failed welds had in common.
  • An instant alert when those conditions appear, telling the team to check the welds on that specific car.
How it ran, in order
  1. 1
    CaptureVoltage, temperature and cycle time from the welding machines
  2. 2
    Map to partsEach weld tied to the part and car being welded
  3. 3
    Map to issuesMatched against past weld failures
  4. 4
    Find patternsVoltage dropping, cycle running short
  5. 5
    Alert per carWhile it is still in the plant
Nothing could be alerted on until the capture existed. The data had to be kept before it could be read.
The impact

No substandard body went out.

The plant can now point to the conditions of every weld, not only to the samples it broke. Bodies with suspect welds are caught and checked before they leave.

What we will measure with the clientFigures to be confirmed
Alerts raised per shiftShows the system is watching
Alerts that found a real weld issueA useful alert, not noise
Bodies held, reworked or cleared after an alertWhat the alert changed on the line
Weld issues found after the plant, before and afterThe proof behind "no substandard body went out"
Robotic arm over an automated line
Automated body line
Yellow industrial robot arm
Every station, every weld recorded
Make every part on your line tell you its story.A short call, or record a message. A person replies within one business day.
Contact usCertified Services Partner, Claude Partner Network