Right first time: Using SPC to improve RFT

What is RFT?

RFT = right first time

In this post, it means that when you start production you get it right the first time, without second try adjustments.

At the beginning of production, we sample the output and send it to the lab to confirm that all the relevant parameter values comply with the spec tolerances.

If everything is OK, we can continue production.

Meanwhile, in most cases, the machine will wait for the lab results. If the machine stops, we are losing output until we get an OK from the lab. If we take the risk of continuing production and the lab results show that we are out of spec, we might get a lot of scraps. If the results are negative, we will need to make an adjustment to get better results and try again and again until we get good results.

RFT – build the process so that you get positive results on every batch.

For that we need to control:

1. the parameters we enter into the machine (speed, temp, etc.)

2. the right % of each element in the formula.

The parameters we enter into the machine – We can use a set of best practice parameters that we use as a constant every time we start production. The parameters are in the machine memory or downloaded from the servers before production. That way we are not dependent on each machine operator who uses different parameters. The operator can adjust the parameters after the first try. If we find better parameters, we adjust their values so that all operators can use the change.

The right % of each element in the formula –

If you are not in the health industry, you can find many formulas not to be balanced enough or having the spec defined badly so that the tolerance is not where it should be. The R&D might define the tolerance to be very narrow without any justifications (meaning that the upper limit and the lower limit are too close to each other). Another possible problem is that the formula is not accurate enough and needs to be adjusted for almost every batch.

Imagine that we have many machines that run many batches. Every batch can have dozens of spec parameters.

How can we find the right parameter that needs to be fixed?

We will find the right parameter to adjust the formula or the machine by using the data we save on every batch. It is almost impossible to do it manually, so we will develop an easy to use tool for that.

But first, we need to understand what is SPC.

What is SPC?

SPC = statistical process control

SPC is a very powerful visual tool to monitor a spec parameter in a process that repeats itself. The idea is to monitor one parameter and see how it changes over time between the upper and lower limits. When we see it approach either limit, we will adjust so that we never get out of spec material.

Example: Suppose that we make chocolate. We want to produce a chocolate bar that contains milk chocolate, nutty cream, and 10 more ingredients. The formula is given to us by the R&D (that means we know the % of each ingredient).

Let’s assume that we want to monitor the sweetness. We will send a sample to the lab at the beginning of the batch and another sample every few hours (according to QA standards). Suppose that this parameter should be between 25 and 32. To reach the spec value we can play with the ratio between the nutty cream and the chocolate milk. Adding more of one ingredient will result in moving the sweetness value up or down.

So let us monitor the production in a visual SPC chart that will show us when we are about to reach the limit and we will adjust before we get out of spec.

In this example the fix was done only when the lower limit was breached, that is the first time that someone noticed the problem. It also means that some of the material/products had become scrap since we got out of spec. That consumes time and materials which are now wasted.

If we use the visual chart, we will see that at the 10th sample we already knew that we were getting very close to the lower limit. The formula fix at the 10th sample would not have created the waste that we got at the 13th sample.

So this is the regular SPC tool that works in most organizations that use samples as part of the process.

The problem:  we usually have many products that have different parameters and different running tools. We can easily produce hundreds or even thousands of data points and it is very difficult to see where we have a problematic parameter and in which product.

Until now this has been classic SPC

How is that relevant to RFT?

In the chart above we saw a good first sample, but in real life, the chart usually looks like this:

We can see that we started out of bound. The first sample is lower than the lower bound. We fix it and start production with a better formula. At this time no one changes the constant formula in the database, because there could be many reasons why this specific sample got a low value and it needs a person who knows what happened to approve the change.

Why there may be no change:

1. During ordinary production time, we don’t see the pattern that returns in many batches. The time between one batch and the next (from the same product) could be weeks or even months.
2. We have a different operator.
3. We might even have a different tool running the next batch.

Each time we start with the wrong parameter we waste time and create scrap.

RFT = be accurate on first try. We want to start with the right parameters to be within and not close to any bounds.

Look at the chart above and you’ll see that the value always goes in one direction (down). What if we adjusted the formula so that we’d start at value 29 ?

1. We wouldn’t have produced scrap at the beginning of the process.
2. It would take longer until we would need to fix the active formula and adjust the parameters (since we started at a higher parameter value).
3. If we start right 90-95% of the time, we don’t need to wait for the lab results at the beginning of each batch since we will have positive results in 9 out of 10 batches. The saving of time and lower scrap will result in a lot of easy to do saving.

We can adjust the formula to reach 29 or even 30.

What is the big problem? How to find the parameters that need fixing?

How can we find the parameter to fix if we have hundreds of products and each of them has dozens of parameters?

How can we find the parameter for sweetness in this product and deal with it?

Find the right parameter to fix

The basic idea:

Let’s take all the first sample results from last year from our production/QA database. For every parameter, we will divide the distance between the upper bound and the lower bound to 4. We will get 3 new boundaries between the upper boundary and the lower boundary (SPEC).

The first line(50%) – in the middle between the upper boundary and the lower boundary.

The second line (25%) – between the lower boundary and the first line (50%)

The third line – (75%) – in the middle between the first line (50%) and the upper boundary.

Now we will look by-product for every sample if the high % of the first samples was above 75% or below 25%. The higher the percentage, the bigger the problem is with this parameter. Let us look at the issue graphically:

The chart shows only the first sample of every batch of the same product. We can see that many times we don’t start within the borders and even when we do, we are very close to the borders, so we need to adjust the formula.

Algorithm: Easily find the parameters that need fixing

The algorithm is very technical, so I will attach an excel file that shows the whole process. I strongly suggest that you use a database application such as Microsoft Access or SQL server or an SQL database. The reason is that you will deal with a lot of data. Databases will make it easier than working in excel, but excel can also work just as well. If you encounter problems, send me an email and I will help you.

Creating the data table

1. Create a table with the following fields: Part (or part number), Parameter, Low bound of spec, Value of the first sample (for the parameter), Upper bound of spec
   
2. Add 2 calculated fields tilted 25% and 75%:
   1. 25% field = Lower bound  + ( (Upper bound – Lower bound)/4)
   2. 75% field = Upper bound – ( (Upper bound – Lower bound)/4)
3. Add another calculated field: “Inbound”
   1. Inbound formula = If sample value is greater than 25% and lower than 75% then inbound = 1, else inbound = 0

Analyzing the data

1. Summarize the Inbound field for each part+parameter and divide it by the number of samples (number of batches).
2. If the value is under 70%, there might be a problem with the parameter at that product.
3. Show the parameter values on the SPC chart and check if the problem is with high or low values.
4. Go back to the developer of the product and ask them to change the formula so that the parameter will get the values in the middle.
5. If other samples (not the first one) always show a decrease or increase in value, then the initial value that we want to receive is close to the opposite bound.
   

 

You can see the whole algorithm in the excel file I am attaching:

The file allows you to enter real data (the pink columns) and the rest is automatically calculated.

After downloading the file do the following:

1. enter your data in the pink columns of the Input sheet. The blue columns are locked and calculated automatically. Erase existing data in the pink columns before entering new data.
2. Go to the Output sheet and hit the refresh all button so that the pivots and the charts will get updated.
3. The pivot will show you the parameters that you should improve.
4. By choosing the Part and the parameter in the upper left corner of the chart you will see how this parameter behaves at the beginning of all batches (shows only the first sample value for each batch).
5. You might need to adjust the axis.
6. One look at the chart and it is easy to see that we need to fix the formula or the spec. After fixing we should see a chart that is more likely to be in the middle.
7. The file is built for 200 rows. If you need more:
   1. add the lines in the pink column
   2. remove sheet protection (no password)
   3. copy and paste the blue cells so that it spreads to the number of records you have
   4. change the source of the pivot table so that it includes the new lines (do the same for the chart).
   5. if you have any problem, contact me at theplanningmaster@gmail.com.

As always, I will be pleased if you press the like button, so I will know that you liked this post and I will write more.

Any remarks or comments feel free to send to:

ThePlanningMaster@gmail.com