What the client was solving for

The client was a contract injection moulder running consumer-goods packaging — closures, containers, and a small thin-wall range. They had a contracted order book exceeding their internal estimate of installed capacity. The natural next step looked like another machine. The CFO asked a sharper question: "Are we sure we are using the four we have?"

The honest answer was: "We do not actually know." Production reporting came from end-of-shift WhatsApp messages. Scrap was estimated weekly from material reconciliation. OEE existed as a number in a spreadsheet that nobody fully trusted.

The operating constraints that shaped the scope

  • Live production cell — any retrofit had to happen without disrupting current orders. No long shutdowns available.
  • Mixed machine generation — two 2018 imported machines with modern controllers, one 2009 machine on an obsolete PLC, one 2014 machine in the middle. Three different control protocols on one shop floor.
  • Multi-SKU operation — typical week ran 6–9 different products across the four machines with frequent changeovers. Cycle-time benchmarking needed to be per-mould, not per-machine.
  • Maintenance team capability — strong on mechanicals, less familiar with sensor and gateway technology. The retrofit had to be supportable by them after handover.
  • Open data — non-negotiable. The client had been bitten before by a vendor platform that locked them out of their own data when the subscription lapsed.

What was actually installed

A deliberately small, non-intrusive Tier 1 visibility layer. No change to control logic on any machine. The line kept running through the install.

  • Industrial PC gateway per machine (4 units), reading directly from existing controller-level signals where available, or via shot-counter and reject-counter signals where not.
  • Current clamps on the main drive and barrel-heater circuits of all four machines — for energy baseline and motor-load fingerprinting.
  • Mould-ID input — operator one-tap on a small tablet at shift start and changeover, so cycle-time data could be attributed correctly to mould (not just machine).
  • One ambient temperature and humidity probe per cell for correlating quality issues against environment.
  • Shared dashboard in the browser, on the shop-floor TV and accessible from the supervisor's phone — open CSV / SQL export, no platform lock-in.
  • WhatsApp alerting for stoppages exceeding configurable thresholds.
  • Five-day operator and supervisor training on what the data meant and what to do with it.

Total install effort: 3 weeks elapsed, 5 cumulative on-site days. Nothing else on the line touched. Existing PLCs untouched. Existing SCADA (where present) untouched.

Cost summary, 2026 ZAR — Hardware (gateways, sensors, clamps, tablets, TV): ZAR ~120 000. Software setup, dashboard, integration, training: ZAR ~160 000. Total project: ZAR ~280 000, single line of four machines. No ongoing subscription required.

What the data exposed in the first 60 days

Visibility almost always reveals problems people sense but cannot name. In this case three things landed in the first two months that justified the entire retrofit several times over.

  • Scrap on two specific SKUs was running at 11–14%, against a plant-wide assumed figure of 4%. Root cause turned out to be one mould with a wear-induced sprue issue and one SKU with an under-spec hold time that operators had been quietly compensating for with longer cycle. Both were fixed inside three weeks.
  • Night-shift cycle times drifted up by 6–9% from 02:00 onward, consistently. The dashboard surfaced it as a per-shift histogram. The cause was barrel-temperature setpoint drift on the oldest machine compounded by an HVAC issue in the cell. Fixing the heater PID and adjusting cell airflow recovered the night-shift output.
  • One specific mould was running 14% slower than its theoretical cycle across all four machines. It was scheduled for refurbishment based on the data. Throughput on that SKU recovered to nameplate after rework.

Measurable outcome

OEE

Plant-wide OEE on the four-machine cell rose by ~17 percentage points over the first 90 days — from a true baseline of 53% (measured, not assumed) to 70% sustained. The "assumed" OEE before the retrofit had been 68%.

Scrap

Cell-wide scrap dropped from ~7.5% to ~3.1% within 60 days, driven entirely by the two SKU-specific findings. No new equipment involved.

Capex deferred

The proposed fifth machine — quoted at USD 320 000 landed — was deferred indefinitely. Headroom on the existing four machines covered the order book at the new OEE level.

Payback

Conservative payback on the ZAR 280 000 retrofit: under four months. Less conservative (counting deferred capex): under one.

What a buyer should take from this case

The fastest path to "more capacity" is almost never a new machine. It is honest measurement of the machines you have. Most factories assume an OEE that is 10–20 percentage points higher than reality. When the gap closes, the next-machine question often disappears.

The second lesson is to start small. A Tier 1 visibility retrofit is the cheapest, lowest-risk way to find out whether bigger capex is justified. Skipping that step and going straight to a new machine — or a Tier 3 controls overhaul — is how plants spend millions to solve a problem they could have diagnosed for hundreds of thousands.

Decision rule: if your true OEE is unknown, do not buy another machine, another shift, or another factory. Measure first. The data will reframe the question, often dramatically.

How CISH structured the engagement

This was a Line Upgrade & Digitalisation engagement using the addanode IoT and OEE platform on the dashboard layer — open data formats, CSV/SQL export, no lock-in. Discussion format on this case is an anonymised metrics summary; production data and exact configuration are reserved for NDA reference walkthroughs.

Related reading: What does it cost to digitalise a production line? and Plastics & packaging production lines.

Frequently asked questions

Does this approach apply to blow moulding and extrusion as well?

Yes. The cycle-time / scrap / changeover / shift-drift pattern is universal in plastics processing. The specific signals differ — barrel zones, melt pressure, takeoff speed — but the visibility-first principle holds.

What if a machine has no usable controller-level signal?

Shot counter, reject counter, and current-clamp data are sufficient for a Tier 1 dashboard. We add controller-level integration where the machine supports it and the customer wants the depth; without it we still get a meaningful OEE picture.

Why no AI / predictive maintenance from day one?

Because real predictive value needs 6–12 months of sensor history. Selling "AI predictive maintenance" without that history is sales theatre. We build the data history first; only then do we layer condition-monitoring or predictive logic on signals that justify it.

Can the dashboard be cancelled without losing data?

Yes — explicitly. The client owns the database. CSV / SQL export is built in. If the service contract ends, the data and the dashboard configuration go with the client.

Can CISH share the named client?

Not publicly. NDA reference walkthrough available where a live buying decision justifies it — including direct contact with the client's production lead.