Skip to content
Planning & productionElectronics & SensorsMachinery & Plant EngineeringMetal & MachiningTool & Mould Making+1

Detect tool breakage by sound

demi listens to presses and stamping machines, recognises punch breakage, misfeeds or missing lubrication by sound and reports them at once.

Today

When a piercing punch breaks, the press keeps running and makes scrap, stroke after stroke, until someone hears it or the parts are caught at the next visual check. At a hundred strokes a minute and more, that quickly means thousands of parts to sort, and the progressive die takes further damage. Only experienced setters know how a healthy press sounds, and they are not at every machine.

An example

On stamping press SA 3 at Rabenhammer Präzisionsteile, bracket H-2207 made of 2 mm sheet runs in progressive die FV-118 at 160 strokes a minute. At 2:12 pm the sound changes during piercing: the agent recognises the pattern of a punch breakage and alerts the shift after four strokes, with stroke count 1,284,406 and container GB-17. The setter stops the press and finds the broken Ø 8 piercing punch; only the 85 parts made since the break are blocked, instead of up to 2,400 parts until the next visual check. The toolroom confirms the finding, and the case sharpens the pattern for every progressive die.
Fictional company, typical workflow.

Your data, wherever it lives

How your presses sound is something only experienced setters hear today; stroke count and die sit in the control and the MES, toolroom repairs in orders and spreadsheets, notes from the shift in the chat. demi reads, understands and links structure-borne sound, strokes, orders and die history, with your rights. The AI recognises a punch breakage after a few strokes instead of at the next visual check, and the shift decides whether to stop and what to sort.

Where the data lives today

  • Photos, video & audio
  • Machines & sensors
  • MES
  • Tickets
  • Office files
  • Chats
Data LayerOne view of everything, with your rights

How demi solves it

Three layers in this order: connect the data first, then the rights, then the AI.

  1. 1 · ConnectData Layer

    Data sources in this case: structure-borne sound sensors on the press frame or die, stroke counters and signals from the press control, orders, dies and lots from the MES and the toolroom's repair history. demi replaces nothing and links every sound event with press, die, order, stroke count and container. So for each die there is a record of how it normally sounds and what earlier breakages sounded like.

  2. 2 · PermitPermission Layer

    Production, toolroom and quality see the sound data and alerts for their presses. Structure-borne sound sensors listen to the steel, not to the people in the hall; conversations are neither recorded nor analysed. Blocking parts in the MES and pulling dies stays with the shift: the agent proposes the block, the shift supervisor confirms it.

  3. 3 · ActIntelligence Layer

    The agent learns the normal sound signature of every die on every press, depending on stroke rate, material and sheet thickness. When the sound deviates, a pattern recognition model matches the deviation to known fault patterns such as punch breakage, slug pulling, misfeed, double blank or missing lubrication. The language model only writes the alert; every confirmation from the toolroom sharpens the pattern for all dies.

How it runs

  1. 1

    Structure-borne sound sensors on presses and dies capture the sound of every stroke; the control supplies stroke count and die.

  2. 2

    The agent compares every stroke with this die's normal sound signature at this stroke rate.

  3. 3

    If it recognises a fault pattern, it alerts the shift at once, with time, stroke count, order and container.

  4. 4

    The shift stops the press, checks the die and, on the agent's proposal, blocks only the parts made since the event.

  5. 5

    The toolroom confirms the finding, and the pattern becomes more precise for every die.

What changes

  • Tool breakage reported after a few strokes, not after thousands

  • Only the parts made since the event are sorted

  • Less follow-on damage to the die

What does it bring you?

Check it with your own numbers

Every starting value is an assumption you can change. The calculation sits right below, step by step.

Assumption, editable

Your numbers

Result

+144 hfreed per year

Assumptions and calculation

demi listens to the presses and reports tool breakage at once; you check and sort only the affected parts.

2 h → 1.4 h per case

How does it run today?

What is already digital saves less: the factor lowers the share demi brings.

How we calculate

20 tool faults × 2 h × 30% × 1.0 (manual) × 12 months = 144 h

Value per year, before demi costs

Freed per year+144 h

Assumption: labour cost of €49.50 per hour in German manufacturing in 2025 (Destatis), plus workplace, IT and overheads.

= Value per year, before demi costs

+€9,360

gained per year

144 h ÷ 1,760 h per role = 0.1 full-time roles

An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.

Combine with more cases and get the package

Questions about this use case

Does demi need new sensors on the press to listen?

Where your presses already have structure-borne sound or force sensors, demi uses their signals. Otherwise one retrofitted structure-borne sound sensor per press, on the frame or the die, is enough, with a small measuring box at the press and no change to the control. Stroke count and die come from the control and the MES.

How does demi learn what a healthy die sounds like?

From the first production runs of each die on each press, separated by stroke rate, material and sheet thickness. For a new die, the agent starts with general patterns and marks its alerts as provisional until enough strokes are recorded. Every finding the toolroom confirms makes detection more precise.

Does demi stop the press itself?

No. demi alerts the shift and the setter within seconds; whether the press stops is decided by them or by your press's existing die protection. The agent proposes blocking the affected parts in the MES, and the shift supervisor confirms it.

Does the system record conversations in the hall?

No. Structure-borne sound sensors sit on the steel and pick up machine vibration, not speech. Only machine signals are stored and analysed, per press and die, not per person.

See this case with your data

In the demo we show you the workflow on your own systems.

Your configurator

No use cases yet

Your configurator is still empty

Add a use case from any card or case page and see what it brings per year, with the calculation. Free, no sign-up, no commitment: try it out and change any number.

Browse use cases