The robot is the visible part and rarely the difficult one. Getting parts presented in a known orientation, at the rate the line demands, from a bin where they arrived in no order at all — that is where the schedule goes. So does verifying that each operation actually happened, rather than that a machine was commanded to perform it.
A station that only performs its nominal operation is a demonstration. A station that also detects when the operation did not happen is a production asset.
| Signal | How | What it changes |
|---|---|---|
| Part presentation | Feeding, singulation and orientation before anything is picked | Parts arrive in bulk and in no order; getting them oriented at rate is the usual bottleneck |
| Position verification | Confirming the part is actually where the program assumes | Fixture wear, tolerance variation and a part seated fractionally proud all defeat blind motion |
| The operation itself | Joining, fastening, pressing, dispensing or placing | The part everyone specifies, and rarely the part that limits throughput |
| Operation verification | Proving the operation succeeded rather than that it was commanded | Torque with angle, force with displacement, presence after placement — measured, not assumed |
| Reject handling | Removing a failed unit without stopping the line | A station that halts on every fault destroys availability faster than the faults themselves |
| Record | Which parts, which parameters, which unit | Traceability that survives a recall investigation, tied to the serial rather than to the shift |
The sensing is the solved part. These are what determine whether the deployment is still running in year three.
Bowl feeders are tuned to a specific part geometry and become unusable when that geometry changes slightly. Vision-guided picking from a bin removes that rigidity and costs cycle time per pick. Choosing between them is an early architectural decision, and getting it wrong is discovered at commissioning when the rate cannot be met.
A stripped thread reaches target torque perfectly well. So does a cross-threaded fastener, and so does one that bottomed out on debris. Torque measured against angle distinguishes a correct joint from several incorrect ones that produce an identical torque reading — and it is the difference between a traceability record and a comforting number.
Components that individually pass inspection can fail to assemble, because the stack accumulates in a direction nobody modelled. Automation is less forgiving here than a human operator, who adjusts without noticing and without recording that they did.
Adding capability to a fast station buys nothing. Line balancing decides throughput, and a station specified in isolation frequently turns out to be the one everything else waits for.
Detecting a wrong part after installation means rework. Preventing its selection — physical keying, gated tool release, interlocked part presence — removes the failure mode. Detection is what remains where prevention is impossible, not the first choice.
A line running several variants spends real time switching between them. Whether that is minutes or hours is determined by fixture design, program management and how much is adjusted by hand — decisions taken long before the first changeover is attempted.
Any line where a sequence of operations has to be performed identically, verifiably and at rate.
Fastening with full torque-angle records, leak and function test, traceability to the vehicle.
Placement, connector seating, screw fastening and functional test with per-unit records.
Device history records, validated processes and change control on anything touching the operation.
Cycle time and changeover between variants as the binding constraints rather than capability.
Part feeding. The robot and the operation are well understood; presenting parts in a known orientation at the required rate, from bulk, is where the effort concentrates. It is usually discovered at commissioning, when the rate cannot be met with the feeder specified.
Because a stripped thread reaches target torque, and so does a cross-threaded fastener or one that bottomed out on debris. Torque against angle distinguishes a correct joint from several incorrect ones producing an identical torque reading.
Hard automation is faster and cheaper for a single high-volume variant; robots earn their cost where variants change or the line has to be repurposed. The question is how many variants over what product life, and it is commercial before it is technical.
Preventing a wrong action rather than detecting it afterwards — physical keying, gated tool release, interlocked part presence. Detection is what remains where prevention is impossible; it should not be the first choice, because detection means rework.
Each unit carries an identity, and every station writes its parameters against that identity rather than against a shift or a batch. The test is whether a recall investigation can establish which components and which process values went into one specific unit.
Often, and it is usually the right place for it — a defect caught at the station that created it is cheaper than one caught at end of line. The constraint is cycle time: the inspection has to complete inside the station's share of takt.
Verifying the result inline.
SOLUTIONWhere models help and where they do not.
INDUSTRYThe line will not slow down.
PLATFORMImaging chain and PLC handshake.
KNOWLEDGEVerdict inside the cycle.
PRODUCTInference at the station.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.