Value stream mapping with measured machine data: filling in the data boxes

This article is for plant managers, production managers and continuous improvement, lean, industrial engineering and maintenance leads who are planning or revisiting a value stream map in manufacturing. The method is well known. The weak point is usually the numbers: cycle time, availability, changeover time and downtime are often estimated or copied from the routing. This article looks at how to fill the data boxes with measured machine data, and what that changes for the bottleneck and the future state.

What Balane writes about value stream mapping: label where every number comes from

Jonas Höttler of Balane GmbH, the Munich company that develops the FlowVisual process modeling and simulation software, has published Wertstromanalyse: Schritt-für-Schritt-Anleitung mit Beispiel (inkl. Software-Vergleich) (in German). The article walks through seven steps from choosing the product family to the future state, works through an example based on a loan application process and explains the key metrics, from takt time to lead time and flow efficiency. In the section on moving from a paper VSM to a robust simulation model, it sets a rule that many plants lack: “Jede Zykluszeit, jede FPY-Quote, jedes Volumen sollte markiert sein als geschätzt, berechnet oder gemessen.” (“Every cycle time, every FPY rate, every volume should be marked as estimated, calculated or measured.”)

The Balane article provides the method and the path to simulation. This article starts from the “measured” category: how to get measured values for the steps of a production line, which data box fields benefit most, and why priorities can shift as a result. The method stays the same; only the data underneath becomes more solid.

Why estimated values distort the map

A value stream map is often built in a few days: a gemba walk, talks with shift leaders, a look at the routing, a few stopwatch readings. That is enough for material and information flow. For the machine data boxes, four biases keep coming back:

• Cycle time comes from the routing. The nominal cycle time goes on the map. The real cycle time drifts with tool wear, material or reduced speed after an incident, and that drift is not in any master data.
• Availability is a flat rate. The same percentage is often applied to every machine. A machine that regularly waits for material and one that hardly ever stops then look the same on the map.
• Changeover time comes from memory. People quote a good day, without the ramp up to the first good part. We explain how to capture the real changeover in Changeover time: how to really measure it, ramp-up included.
• Micro-stops are invisible. Interruptions of a few seconds to a few minutes never reach an incident log. Added up, they can cost more time than major breakdowns, yet they appear nowhere on the map.

Then there is the sampling effect: on the day of the walk, the line runs under observation, often on the morning shift with the most experienced crew. The night shift and the difficult variants are missing from the sample.

Which values go into which data boxes, and where they come from

For every process step with a machine, it pays to measure over two to four representative weeks, across all shifts. The table shows which fields benefit:

In the spirit of the Balane article, keep the origin visible: every value in a data box is labeled measured, calculated or estimated, with the measurement period. Anyone reading the map then knows how far it can be trusted.

What changes for the bottleneck

Value stream mapping compares the cycle time of each step with customer takt: a step whose cycle time exceeds takt limits the flow. With nominal values, that comparison quickly becomes too optimistic, because a machine only produces for part of the planned time. A more telling figure is the effective time per good part: real cycle time divided by the share of time the machine actually produces.

An illustrative example with fictitious values: customer takt is 60 seconds. According to the routing, step A has a 50 second cycle time and step B 55 seconds. On paper, B is the bottleneck. Measurement then shows that A loses a quarter of its planned time to frequent changeovers and many micro-stops, while B loses only a small share. The effective time per part for A is then above 66 seconds, over takt, while B stays just under 58 seconds. The bottleneck is A, and that is where countermeasures belong.

Shifts like this are not exceptions. They appear whenever losses differ strongly between machines. A map filled with nominal values cannot show them, because it does not contain those losses.

Future state and simulation: start from measured losses

Measured data also change the future state. If the bottleneck comes from micro-stops, root cause analysis and autonomous maintenance help more than an extra buffer. If it comes from changeovers, SMED is the lever. If values vary strongly between shifts, a shared standard is the first step. For a simulation, which the Balane article presents as the next step after the paper map, measured data also provide distributions rather than averages: how often a machine stops, and for how long. The approach:

1. Scope the value stream. Set the product family, process steps and customer takt according to the method.
2. Measure the machines without intervening. Two to four weeks across all shifts, with stops and changeovers and their reasons.
3. Fill in and label the data boxes. Mark each value as measured, calculated or estimated, with period and spread.
4. Identify the bottleneck. Compare effective time per good part with takt, not nominal cycle time.
5. Design the future state and measure again. Derive actions from the loss Pareto and keep measuring to check their effect.

Because measurement keeps running, the value stream map also becomes repeatable: a year later, the current state can be mapped again with the same rules. For the traps to avoid along the way, see Performance tracking: 5 measurement pitfalls that distort decisions.

Who brings what: method, simulation and measurement

Method, facilitation and simulation remain the job of the lean team and its tools. TEEPTRAK provides the measured machine data. A module wired into the electrical cabinet and a sensor at the line record the state of each machine automatically, with no connection to the PLC. Installation takes about one hour per machine, and OEE is available within 48 hours. Operators assign a reason to stops, and micro-stops are detected automatically. TEEPTRAK is not an MES: it measures OEE, downtime and micro-stops in more than 700 plants across more than 30 countries.

Further reading

• Wertstromanalyse: Schritt-für-Schritt-Anleitung mit Beispiel (inkl. Software-Vergleich) (Jonas Höttler, Balane GmbH, in German)
• Changeover time: how to really measure it, ramp-up included
• Performance tracking: 5 measurement pitfalls that distort decisions

FAQ

How long should you measure before filling in the data boxes? Long enough to cover all shifts and a representative product mix, usually two to four weeks. A measurement that is too short describes one particular week, not normal operation.

Does automatic measurement replace the gemba walk? No. Walking the floor, talking to teams and observing remain essential to understand flows, inventory and causes. Measurement only replaces the estimated values in the machine data boxes.

What is the difference between cycle time and effective time per good part? Cycle time is how long one part takes while the machine is running. Effective time per good part also includes stops, changeovers and micro-stops. It is the better figure to compare with customer takt.