Rethinking the Definition of Changeover Time in SMED with the Help of Artificial Intelligence

Changeover times play a major role within the losses of equipment effectiveness. The Toyota Production System devotes a significant chapter to reducing changeover losses with the SMED methodology. But how precise can a changeover time be measured or how do we assess the quality of a changeover process? AI can help in the future.

A practicable definition of changeover time is that the changeover time occurs between the last product of the previous lot and the first product of the next lot. If an equipment produces either on 0 or 100% of performance, this definition is valid. However, there are many systems where the unit number curve is different. In practice, there are often run-down and / or start-up losses before or after the actual standstill. A more precise definition of the changeover time would be, for example, to define the time between the last product of the previous lot at full speed up to the subsequent product at full speed as changeover time. But this in operations reality hard to measure as machines often do not run precisely with a stable output e.g. per minute and the expected performance rate is often unequal to the realized rate.

The typical course of a unit number curve before and after a changeover procedure is shown in the following figure.

Image: Unit curve before and after the changeover

If the expectation of the full speed is integrated in the changeover time definition, the three time components of the run-down time, idle time and start-up time blur as the point in time, when the unit number curve is leaving the stable run (= crest line) is not defined sharply. Here artificial intelligence can provide more precision with its classification capabilities of time series data.

Deep Learning can use a trained neural network to classify time series data into stable and unstable conditions. This division forms the basis for a precise decomposition of the time portions of the changeover process. If the system leaves the stable run it is up to the number of pieces “zero” in the run-down. Then the standstill begins during which the mechanical conversion happens. From the beginning of production, until the stable production level is reached again, the period is defined as start-up.

This data-based classification offers several advantages in practice:

  • Even with systems that do not produce an identical number of pieces every minute in full production, the time at which the crest line is left or reached can be clearly defined by the algorithm.
  • The run-down time (or run-down angle), and thus the management of the process, can be precisely measured and improved.
  • The start-up time (or start-up angle) is a measure of the conversion quality and can be precisely measured and improved.
  • Both run-down and start-up losses can be accurately expressed in lost pieces or lost OEE percentage points.

The following presentation visualizes the discussed points.

Image: Unit curve before and after the changeover including run-down and start-up

In practice, the biggest advantage is expected in the ability to measure the conversion quality of the equipment. E.g. a short downtime time followed by a long ramp-up time can be compared with a longer downtime but steeper ramp-up curve. Which of these scenarios is – based on facts – a better result for the Overall Equipment Effectiveness?

This approach is not a solution for advocates of slip-and-stick solutions. However, it’s a viable technology option that creates facts in places that were not measurable without advanced analytics in the past. In this way, one can break away from the discussion about the assessment of the facts and come to the actual execution of the process improvement – and that’s an improvement itself.