4.5 SPC & Dashboards
Real-time trends, alarms,SPC and
howdashboards bring discipline to feed corrections back to print / PnP / reflow without drama.
Statistical Process Control (SPC) keepsthe production quality steadyfloor by monitoringturning liveraw process signalsdata insteadinto clear, real-time guidance. Instead of waiting for defects to appearsurface, later.live Thecharts focustrack isthe onvital asigns fewof keyprinting, metrics—like SPI transfer efficiency, placement miss rates,placement, and reflow peak/TAL—trackedso small drifts are corrected before they escalate. The key is simplicity: a handful of metrics, split by family where it matters, presented in simple,a readable charts soway operators can spotact drifton atwithout ahesitation. glance.With Alarm tiers define exactly what happens when a parameter shifts, from a gentle warning to a production stop, and each alarm isalarms tied to predefined corrections and every change logged against time and product, dashboards become a small,stable predefinedfeedback correctionsystem rather than guesswork.a Metricsnoisy are split by feature family to avoid hiding trouble in averages, and every recipe, stencil, and lot change is time-stamped on the dashboard so cause and effect are clear. By keeping the scope tight, linking signals to immediate actions, and using the same disciplined feedback loop across print, placement, and reflow, SPC dashboards evolve from decorative displays into reliable, low-drama tools for keeping the line in control.
4.5.1 What SPC is (in factory words)
Statistical Process Control = watch the process while it’s running, not just the parts after. You pick a few live signals, chart them simply, and agree what happens when they drift. The goal isn’t pretty graphs—it’s early, small corrections that keep defects from forming.
Think of it like a car:
- Speedometer → SPI transfer efficiency (are we printing the right volume?)
- Fuel light → Mounter starvation/miss rates (are feeders about to bite us?)
- Engine temp → Reflow TAL/peak/ΔT (are we cooking the board right?)
If any gauge moves, you nudge a knob now, not after AOI screams.
4.5.2 The “few, vital” metrics (by station)
Start with these. Add more only if they change decisions.
4.5.3 Simple charts that operators actually read
- I-MR (Individuals & Moving Range) for SPI TE and TAL/peak per panel.
- p-charts for AOI false calls/board.
- Run charts (sparklines) for starvation minutes, misses/retries, wipes/board.
Keep the visuals dumb-clear: traffic lights (green/yellow/red), last 50 panels, and a vertical line labeled “change made” whenever someone tweaks a recipe or cleans optics. If people can’t read it at 2 meters, it’s too fancy.
4.5.4 Alarm tiers (decide once, use everywhere)
Make the MES or line PC enforce this—humans forget when the line is hot.
4.5.5 Closing the loop (the “if this, then that” map)
Tie each dashboard needle to a small, specific action:
- SPI TE low on chips (Cpk <1.0) → reduce blade pressure one notch and shorten clean interval; recheck next 3 panels.
- SPI area creep near fine pitch → slow separation 10–20%; if persists, wet+vac, then inspect apertures at next stencil spin.
- PnP miss spikes on Feeder 12 → verify peel path, run 20 dry picks; if still bad, swap feeder; quarantine reel lot.
- Starvation minutes climbing on constraint → pull next splice earlier (countdown threshold), load spare pre-threaded feeder.
- TAL short on cold TC → slow belt a touch (time first), then re-aim late zone by +3–5 °C.
- ΔT at peak > target → add soak 10–20 s via mid-zone setpoints; tick blowers up one step.
- AOI bridges rising in one region → check printer separation and cleaning; don’t widen AOI until print is fixed.
- AXI QFN voids trending up → verify paste lot/date; lengthen soak slightly; consider N₂ A/B run.
These moves should be printed next to the screen. No guessing.
4.5.6 Roles & views (one dashboard does not fit all)
- Operator view (per station): 3–5 gauges, next-action button (“Add clean now”, “Check Feeder 12”), and clear pass/stop banner.
- Lead / Supervisor: line balance (cycle time per machine), red/yellow counts per hour, top 3 issues to clear this shift.
- QE/PE weekly: Cp/Cpk trend by family, escapes vs AOI/AXI limits, before/after of any recipe or stencil change.
- Management: OEE (Availability/Performance/Quality) plus first-pass yield. No microscope photos.
4.5.7 Make data trustworthy (plumbing that matters)
- Everything gets an ID: recipe revs (printer/PnP/oven), stencil ID, paste lot, feeder IDs, operator badge, WO/panel/SN.
- One time base: all stations sync to the same clock so panel histories align.
- Auto-attach evidence: SPI/AOI/AXI images and reflow plots stored against the panel ID; no screenshots in chat apps.
- Annotate changes: pressing “changed belt speed +5 mm/min” drops a vertical line and a note on charts.
4.5.8 Start small (a rollout that sticks)
- Pick one product and wire SPI TE Cpk, wipes/board, PnP misses, TAL/peak, AOI false calls.
- Agree alarm tiers and the first corrections (Section 10.5.5).
- Run two weeks; collect “we touched X, chart did Y” stories.
- Freeze the layout; clone to the next product with only limit tweaks.
4.5.9 Common traps (and quick escapes)
- Too many charts. Result: no one looks. Trim to the five that drive actions.
- Alarms with no playbook. Result: people mute them. Add the button that does the first fix.
- Averages across families. Result: masked trouble. Always split by feature family (chips/QFN/BGA/thermal).
- Goal-post moving. Result: pretty charts, bad product. Change the process, not the spec, unless customer agrees.
4.5.10 Pocket checklists
Designing the dashboard
- 5–7 metrics total; split by feature family where relevant
- Green/Yellow/Red bands defined; MES enforces Stop tier
- One-click first correction per alarm
Daily use
- Operators clear warns with the buttoned action; notes auto-logged
- Leads review top 3 issues at break; assign quick fixes
- QE samples escapes vs AOI/AXI and feeds back to process if trend
Change control
- Any recipe/stencil change = vertical line + comment on charts
- Weekly Cp/Cpk & Pareto review; only small rule edits, not broad relax
Bottom line:When SPC
isn’tfocuses abouton worshippingthe graphs—it’sfew aboutsignals seeingthat driftmatter, earlyenforces consistent alarm responses, and pressingfeeds theresults smallestback buttoninto process control, it keeps production calm and predictable. The outcome is fewer surprises, steadier yields, and inspection systems that bringsspend less time firefighting and more time confirming that the line backis toalready center.on Keep the metrics few and useful, wire alarms to actions, and log every tweak. Do that, and your dashboards stop being wallpaper; they become quiet autopilots for print, place, and reflow.