How to measure OEE without replacing your machines

The problem on most shop floors

Your production manager knows the press line had a bad week. What nobody can tell you is how much of that was breakdowns, how much was slow running and how much was rejected parts. The numbers live in a register at the machine, a supervisor's shift report and an Excel sheet that someone fills in on Monday morning from memory.

Overall equipment effectiveness, or OEE, gives you one number per machine per shift that splits the loss into those three parts. And you can usually measure it on the machines you already own. You do not need a new CNC or a new moulding machine to start.

Quick answer

• OEE = availability x performance x quality.
• Availability is run time divided by planned production time. It captures breakdowns, changeovers and waiting for material.
• Performance is actual output against the ideal cycle time for the time the machine ran. It captures slow cycles and short stops.
• Quality is good parts divided by total parts. It captures rejects and rework.
• For older machines, add a sensor or tap an existing signal, send it through a small gateway, and have operators enter downtime reasons on a tablet at the machine.

A typical first phase covering 10 to 20 machines takes 4 to 6 weeks, including the time it takes operators to trust the screen.

Getting signals out of old machines

Machines fall into three rough groups, and most plants have all three.

1. Machines with a modern controller. Newer PLCs and CNC controllers often speak OPC UA, Modbus TCP or MTConnect. You read cycle counts, run state and alarms directly. This is the easy group.
2. Machines with a controller but no open protocol. Older PLCs may have a serial port or a vendor protocol. A protocol gateway can often read the tags you need. Check with the machine maker before anyone opens a panel, and never change PLC logic just to get data out.
3. Machines with nothing usable. A press from the 1990s, a manual lathe, a furnace. Here you add sensors: a current transformer on the motor supply to tell running from idle, a proximity sensor or counter on the stroke or ejection, a temperature probe where it matters. These are clamp-on or bolt-on and do not touch the machine's control circuit.

The signals go to an edge gateway, a small industrial computer on the shop floor, which buffers data when the network drops and sends it on, usually over MQTT, to a server or cloud database. Shop floor Wi-Fi is often weak near large machines, so plan for wired Ethernet or a dedicated access point where you can.

As a rough example, sensors and a gateway for a machine with no controller might cost INR 15,000 to 40,000 per machine in hardware, depending on what you measure. The software and integration work is usually the larger part of the budget.

The part that decides whether it works: downtime reasons

A sensor can tell you the machine stopped. It cannot tell you why. Was it a tool change, a material shortage, a quality hold, a breakdown, or the operator at lunch?

That answer has to come from a person, so make it easy.

• Put a tablet or a rugged touchscreen at the machine or the cell.
• When a stop lasts longer than a set time, say three minutes, the screen asks for a reason.
• Keep the list short, eight to twelve reasons, with large buttons and labels in the language the operators speak. In a Gujarat plant that may mean Gujarati and Hindi, not English.
• Let supervisors correct reasons at the end of the shift, with a record of who changed what.

If operators see the reasons used to blame them, they will pick "Other" every time. If they see the reasons used to fix the material supply problem they have complained about for a year, entries get better within weeks.

What to connect it to

OEE on its own is a scorecard. It becomes useful when it sits next to the data that explains it.

• Production plan. Planned production time comes from the shift plan or your ERP, whether that is Tally with a production add-on, SAP Business One or something built in house. Without it, availability numbers are guesses.
• Ideal cycle time. Agree it per part per machine with engineering, and write it down. Using the "usual" cycle time hides performance losses.
• Quality records. Rejects entered at the inspection station should flow back to the machine and shift that made them.
• Maintenance. Breakdown reasons should open a maintenance ticket, so the same data drives both the OEE report and the repair history.

Mistakes and limits

Mistakes we see often

• Starting with the whole plant. Pick one line or one bottleneck machine. Prove the numbers are right there first.
• Arguing about benchmarks. Your first accurate OEE number will probably be lower than you expect. That is normal. Compare each machine with itself month on month.
• Dashboards nobody looks at. Put one large screen on the shop floor showing the current shift, and send a short daily summary to the plant head on WhatsApp or email. A weekly review meeting that uses the top three loss reasons does more than a dozen charts.
• Ignoring the planned time question. Decide early whether tea breaks, planned maintenance and no-order days count as planned time. Different answers give different OEE, and people will compare them.

When you do not need a custom system

If you run a handful of machines and only want run and stop status, an off the shelf IoT kit with its own dashboard may be enough, and cheaper. A custom build starts to pay when you need OEE tied to your own ERP, your own reject codes and your own production plan, or when you run mixed old and new equipment across several plants.

Either way, start small, get the downtime reasons right and let the numbers earn the shop floor's trust before you scale.