As part of a much larger series on Mixed Model Sequencing, this post describes how to verify the sequence quality. It also describes how to determine the required buffer spaces to buffer against these fluctuations in workload. There may be some wiggle room here. Read on:
Verify Sequence Quality
Next, we have to verify the quality of the sequence. Let’s demonstrate it with the sunroof-mounting station. The average cycle time at this station is 52.1 seconds. Whenever a model 2434 with a sunroof comes along, it takes 111 seconds and hence 58.9 seconds longer than average. For a model 2437, it takes 112 seconds and hence 59.9 longer than average. All other models have no sunroof, and they have a cycle time of 0 seconds and hence 52.1 seconds less than average. Please note that the average is based on the average cycle time of this station and not the line takt or target cycle time. The table below shows all the data, with the sunroof mounting over/under times marked in red.
Now we simply go through the sequence and sum up the times that the process is over or under average speed. Across all parts this should give a value of zero again. Below is once more the sequence we had from the last post.
Starting at zero, the first part has a sunroof, adding 59.9 to the accumulated times. The second part had no sunroof,reducing it by 52.1 seconds to 7.9 seconds. The third part added a roof again adding 58.9 seconds to an accumulated total of 66.8 seconds at the third slot. The graph below shows the first 30 slots. At slot 22 we had two “non-sunroof” vehicles in a row. In the long run the values will go up and down, they can also go negative, to eventually reach zero again.
The graph below shows the first 30 slots for all stations. Notice how the sunroof station has the most regular zigzag curve? These were the first two product types we sequenced.
Overall, you will notice an upward trend at most stations. While eventually things will go down again, larger deviations from the average are not good, as this requires larger buffers. Hence, this is not such a good sequence.
The reason for this is part type 2436. You remember how in the last post this part was the last one we sequenced, and it had a pretty bad sequence. While there should have been a part every 3.28 slots, due to other parts being sequenced earlier this part had to do with the slots that were left at the end. Hence, during the first 30 slots there were only 6 of this part type, where there should have been 9.
This is a problem because across the board this part type had a very low work content. The average work content per station for this part type was only 37.2 seconds, much less than the average of 52.04. The data is once again shown below, with the average work contents for all parts shown in red.
Hence, it would have been a much-needed part type to reduce the workload again. Unfortunately, this part type is now very unevenly distributed, and our sequence is not that good. Remember how I mentioned Mixed Model Sequencing being an iterative process? That’s right, we should go back and do the sequence again. Maybe this time we will sequence part 2436 earlier, possibly directly after part types 2437 and 2434 (the two part types with a sunroof).
Also remember how I mentioned in an earlier post that there are many different aspects that can influence sequencing? In our example we used mostly the largest cycle times for any product at at any station, but maybe we should have also been looking at the work content per product variant. Hence, for a real-world sequence I would go back and do it again, hopefully better – unless I find another aspect I should have been considering too. Here, however, I continue with the sequence we already have. Again, it is iterative and more of an art than a science!
Buffer Size
Anyway, let’s have a look at the buffer size. The table here shows the largest positive and negative accumulation of work content (i.e., the value of the top and bottom most peaks).
We are interested in the spread (i.e., the difference between the smallest and the largest peak). This is the fluctuation that we have to cover. This is the amount of buffer that we would need to provide for a smooth operation.
For a continuously moving assembly line, this would be represented by the width of the space allocated to this workstation. If your line moves 6 meters per minute, then a buffer of 72.7 seconds is an additional 7 meters of slot space.
For a line with buffer slots between stations, this spread would have to be divided by the cycle time (or takt time, depending what you used to calculate the spread) to determine the number of buffer spaces.
Looking at the data table above this seems like a lot, but there is some wiggle room. This peak buffer space is needed only a very few times during the production. Here are a few options:
- You probably plan to include some buffer anyway. This buffer here does not have to be on top of the regular buffer, but can be combined with the regular buffer. For example, your average cycle time for the rear seats is 51.7 seconds. If the whole line moves at a line takt of 60 seconds, then you automatically have 8.3 seconds buffer here. If the work content fluctuation in you rear seat station, ask for a buffer of 49.7 seconds, then you do not have to add them, but merely take the larger one.
- Humans work faster if there is a lot of work, and slower if there is less. Hence, a part of the fluctuation can be buffered with a changing human work speed. This can be 10% of the buffer, or 20% if you’re daring, but probably not 30%. The tighter the buffer, the more likely that the workers will not be able to make it even with a (short time) extra effort.
- But even this is not a huge problem. If the buffer is not enough, then the station cannot make it in time, and the rest of the line has to wait. If this happens once per shift for 10 seconds, so what. While not ideal, a delay for the rest of the line of 10 seconds in a shift may be preferable to an extra 3-meter floor space for buffer all the time!
So you see, there is some wiggle room for the buffer. If it is not wiggly enough for you, you could look for an even better sequence. Or you can also change the product design, add better machines, and do a lot of other things as discussed in a previous post to make these problems simply go away. Or you bite the bullet and add the buffer. Or you try if you can get away without that much buffer, and change it back if it does not work. The possibilities here are endless. Now, go out, make your buffer fit your needs, and organize your industry!
P.S. Many thanks to Mark Warren for his input.
Series Overview
- Mixed Model Sequencing – Introduction
- Mixed Model Sequencing – Just Make the Problem Go Away
- Mixed Model Sequencing – Adjust Capacity
- Mixed Model Sequencing – Basic Example Introduction
- Mixed Model Sequencing – Basic Example Workload and Buffering
- Mixed Model Sequencing – Basic Example Sequencing
- Mixed Model Sequencing – Complex Example Introduction
- Mixed Model Sequencing – Complex Example Data Basis
- Mixed Model Sequencing – Complex Example Sequencing 1
- Mixed Model Sequencing – Complex Example Sequencing 2
- Mixed Model Sequencing – Complex Example Verification
- Mixed Model Sequencing – Summary
Here is also the Sequencing Example Excel File for posts 7 to 11 with the complex example. Please note that this is not a tool, but merely some of my calculations for your information.