Rollout and data8 min read

When to Connect a Digital Twin to Live Data, and When Static Data Is Enough

DBR77 TeamPublished

Connect a digital twin to live machine data when the team uses the model often, weekly or daily, and wants it to follow how the line runs from week to week. Stay with static data when the decision comes once or twice a year: a new layout, a robot cell, a new hall, a capacity investment. Most first projects belong to the second group. Data the plant already has answers them: ERP, routings, nominal or measured cycle times. Live data can follow later, once the model is in regular use, as an individual integration project.

What "static" and "live" mean in a factory model

Static data is loaded into the model at a point in time. It includes the CAD layout, routings, cycle times from ERP or a stopwatch, changeover times, transport speeds, the shift calendar and the order mix. When the plant changes, someone updates the model by hand.

Live data comes from the shop floor through an integration built for your plant: machine states, stop reasons, cycle counts and actual cycle times, for example from a system that already collects them. The twin does not connect to PLCs or robots itself, and every integration is an individual project. Once the feed is in place, the model's parameters stay close to what the line does today.

Static data is enough for a decision when the base model matches the plant as it runs today. The model owner updates that base when the plant changes, for example after a new machine or a new shift pattern, and each option is tested as a variant on a copy of it. A static model still shows what averages hide: queues, shared machines, order sequence and due dates, because it simulates the week over time. See Averages vs simulation over time.

Why "integration first" often delays the value

Many projects plan to connect every machine before the first simulation runs. Signals must be mapped, cleaned and time-stamped, and old machines need extra hardware. This work takes months.

Meanwhile, the factory keeps deciding. The layout gets chosen on a spreadsheet, and the robot gets ordered on the supplier's cycle time.

A second problem is that live data cannot repair a weak model. If the routing logic is wrong, or the buffer rules do not match the floor, a live feed only makes the wrong result look fresh. The model logic has to be right first, and that is easier to check with static data that engineers can read line by line. Even nominal cycle times from routings are a useful start: comparing the model's output with what the floor actually produces shows the gap between theory and reality, and that gap points to the real bottleneck. We describe the minimum data set in What Data Do You Need to Start a Factory Simulation?.

Five questions before you connect anything

Answer these in writing before you plan an integration. If most answers fall in the left column, static data is enough for now.

QuestionStatic data is enough whenConnect live data when
How often do you decide from the model?A few times a year, at investment or layout gatesWeekly or daily, for planning or picking
How fast do the inputs change?Cycle times and routings stay stable for monthsCycle times drift, new products and changeovers come every week
What does a stale input cost?A small error does not change which variant winsA stale input changes the plan for the next shift
Who uses the result?Managers in a review meetingA shift lead or planner who acts on it the same day
Can you own the data feed?Nobody has time to watch data qualityA named person fixes missing or wrong signals

Most teams skip the last question. A live feed needs an owner and a rule for what happens when data arrives late.

What static data can already answer

Two DBR77 case studies show how much a model built on static data delivers.

A plastic injection molding company for the household appliances industry wanted to automate several stations on a line. DBR77 built a twin of part of the plant from material flows and cycle times per task. The model found the bottleneck at a manual step done by the first operator: 90 seconds of activity and 98 seconds of waiting. No live feed was needed to see that.

An HVAC manufacturer with an average of 800 picks per shift needed a better order for withdrawing material from the warehouse. The simulation was built from current production orders. To update it, the team only imports revised production orders, and the algorithm finds a new solution in seconds. That is useful when material or people are short.

The second case shows a middle path. The model is not connected to machines, but a simple import refreshes it whenever the plan changes. For many plants, this covers daily needs.

See what a model built on static data can answer

The automation concept and the warehouse picking case above both ran on data the plants already had. The use cases list more projects of that kind, from layout variants to intralogistics.

Browse use cases

When live data earns its place

Live data is the next stage of working with the model, after the static version has shown the team where the real bottleneck is. It pays off in three situations.

  • The decisions recur. When the team plans the product mix or shifts with the model every week, each run should start from how the line actually ran last week. Live data keeps the model's cycle times in line with the floor between those runs.
  • The inputs drift. Cycle times creep up, a machine runs slower, a new tool changes changeover times, and live data shows the change before the next review meeting does.
  • A change has just gone live. After a new layout or robot starts, live data shows whether the line behaves as the model said. If it does not, you learn whether the model or the floor needs fixing.

A twin is built to test changes, and an alarm system for the shift is a separate tool. If alerts are your goal, read Why IoT Alerts Fail on the Shop Floor and What Works Instead. Where the data is processed is a separate architecture question, covered in Edge vs Cloud in Manufacturing.

Connect in steps, not all at once

A practical path has four steps.

  1. Start static, for one decision. Build the model from layout and process data and use it to compare variants of one real change.
  2. Calibrate with history. Compare the model's output with what the line actually produced. The gap between the two shows where theory and the floor differ, and that is usually where the real bottleneck sits. Replace estimated cycle times with values from logs, ERP or historical machine data, and check the result with the people who run the line. They know what a bad week looks like.
  3. Connect the few signals that change a result. Usually these are the states and cycle times of the bottleneck and the stations around it. In an illustrative plant with 40 machines, three or four stations near the bottleneck may decide the output. Connect those before any other sensor in the hall.
  4. Set owners and failure rules. Decide who fixes a missing feed, and what the model uses when data is late, for example the last approved values.

A model nobody is responsible for stops matching the plant after the first change on the floor, whether it is fed by hand or by sensors. We cover roles in Who Should Own the Digital Twin Model?.

FAQ

Does a digital twin need real-time data to be a digital twin?

No. A digital twin is a model of your plant that runs production over time. When the plant changes, for example with a new machine or a new shift pattern, the model owner updates it from new data or measurements, and that version is the base for the next decision. Live data is a later stage, when the model is used regularly and should follow the plant between those changes.

Is a model built on static data good enough for an investment decision?

Yes, if the inputs are measured, not guessed. Layout and investment decisions need a fair comparison of variants on the same inputs, and static data from routings and cycle times supports that.

Which machine data should we connect first?

Start with the bottleneck and the stations next to it: machine states, stop reasons and actual cycle times. These signals move the result most. The rest can wait.

What happens when a live data feed fails?

Agree on it before you connect. A common rule is that the model falls back to the last approved values and flags the gap, so nobody plans on missing or wrong data.

Conclusion

Take the next decision on your list and ask how often it will come back. If it is a once-a-year investment, build the model now from the layout, routings and measured cycle times. If the team will rerun it every week, name the person who will own a live feed from the bottleneck stations before anyone orders hardware.

Start with the data you already have

Your layout drawing, routings and measured cycle times are a workable starting point. In the demo you see a model built from that kind of input, and how live machine data could be connected later as an individual integration.

Talk to an expert

Sources

Want to test a decision from your plant?

Describe the decision to a DBR77 expert and we will show how DBR77 Digital Twin compares the options on output, waiting time and transport load.

Talk to an expert

We use your data only to answer your request. See our Privacy Policy.

Digital Twin newsletter

Short notes on testing layout and flow decisions on a model, and news from DBR77 Digital Twin. You can unsubscribe at any time.