Maximizing the exploitation of critical Capacity Constraint Resources (CCRs), so called bottlenecks, is crucial for maximizing revenue. Changeovers usually have a significant impact on productive capacity, reducing it with every new change made on those resources that already have too few of it.
Yet changeovers are a necessary evil, and the trend is going for more frequent, shorter production runs of different products, so called high mix / low volume. Consequently, changeovers must be kept as short as possible in order to avoid wasting the precious limited productive capacity of Capacity Constrained Resources (CCRs).
Monitoring changeover durations at bottlenecks is a means to:
- reinforce management’s attention to the appropriate CCR management
- analyze current ways of changing over
- improving and reducing changeovers duration
Management’s obsession should be about maximizing Throughput of the constraints.
To learn more about this, read my post “If making money is your Goal, Throughput is your obsession”.
What data for changeover monitoring?
When starting to have a closer look at how capacity is lost during changeovers, the question is: besides direct periodic observations, what data are necessary and meaningful for such monitoring?
Before rushing into a data collecting craze, here are a few things to take into account:
In the era of big data, it is admitted now that one never has enough data. Yet data must be collected somewhere and possibly by someone. The pitfall here is to overburden operators with data collection at the expense of their normal tasks.
I remember a workshop manager so passionate with data analysis that he had his teams spent more time collecting data than to run their business.
Chances are that your data collection will be manual, by people on shop floor. Keep it as simple and as short as possible.
This a matter of respect for people and a way to insure data capture will be done properly and consistently. The more complicated and boring the chore, the more chances people will find ways to escape it.
Take time to think about the future use of data, which will give you hints about the kind of information you need to collect.
Don’t go for collecting everything. Essential fews are better than trivial many!
Be smart: don’t ask for data that can be computed from other data, e.g. the day of the week can be computed from the date, no need to capture it.
Example of data (collected and computed)
- Line or machine number
- Date (computed)
- Week number (computed)
- Changeover starting date and hour
- Changeover ending date and hour
- Changeover duration (computed)
- Changeover type
- Shift (team) id.
Explain why you need these data, what for and how long presumably you will ask for data capture. Make yourself a note to give data collectors regular feedback in order to keep people interested or at least informed about the use of the data.
Data relative to resources with significant excess productive capacity can be ignored for the sake of simplicity and avoid overburdening data collectors. Yet chances are that some day you’ll regret not having captured those data as well, and soon enough. Make your own mind about this.
Monitoring: what kind of surveys and analyses?
There are roughly two types of analyses you should be looking for: trends and correlations. Trends are timely evolutions and correlations are patterns involving several parameters.
One key trend to follow-up is changeover duration over time.
Monitoring by itself usually leads to some improvement, as nobody wants to take blame for poor performance i.e. excessive duration. As frequently things tend to improve spontaneously as soon as measurement is put in place, I use to say measurement is the first improvement step.
The first measurements set the crime scene, or original benchmark if you will. Progress will be appraised by comparing actual data against the original ones, and later the reference will shift to the best sustained performance.
In order to compare meaningful data, make sure the data sets are comparable. For instance certain changeovers may require additional specific tasks and operations. You may therefore have to define categories of changeovers, like “simple”, “complex”, “light”, “heavy”, etc.
Over time the trendline must show a steady decrease of changeover durations, as improvement efforts pay off. The trendline should fall quickly, then slow down and finally reach a plateau* as a result of improvements being increasingly difficult (and costly) to achieve, until a breakthrough opens new perspectives: a new tool, simplified tightenings, another organisation…
*See my post Improving 50% is easy, improving 5% is difficult
>Consider SMED techniques to recover capacity
Looking for correlations is looking for some patterns. Here are some examples of what to check:
Is there a more favorable or unfavorable day of the week? If yes, understanding the cause(s) behind this good or poor performance can lead to a solution to improve everyday performance.
Does one team outperform underperform? Is one team especially (un)successful? The successful team may have better practices than the lower performing ones. Can those be shared and standardized?
For instance if one team consistently outperforms, it could be this team found a way to better organize and control the changeover.
If it is the case, this good practice should be shared or even become the standard as it proved more efficient.
I happen to see the performance data from a night shift in a pharma plant being significantly better than the day shifts. Fewer disturbances during the night was the alleged cause.
Be critical: an outstanding team may “cut corners” to save time. Make sure that all mandatory operations are executed. Bad habits or bad practices should be eradicated.
Conversely, poor performing teams may need to be retrained and/or need coaching.
Is one type of changeover more difficult to master? Search for causes and influencing factors. Some engineering may be required to help improving.
These are only some examples of patterns that can be checked. Take time to consider what factor can have some influence on changeover ease and speed, than check how to test it with data and how to collect these data.
Note that correlation is not causation. When finding a pattern, check in depth to validate or invalidate your assumptions!
Speak with data
All the data collection and analysing is meant to allow you and your teams to speak with data, conduct experiments in a scientific way and ultimately base your decisions on facts, not beliefs or vague intuitions.