The Application of JMP in the Six Sigma Methodology: Practical Examples

Good afternoon. I'm Marjoleine Wevers from Robert Bosch. Today I will explain the application of JMP in the Six Sigma methodology with practical examples together with Elodie Delclaux.

I'm part of the Bosch Group. The Bosch Group everyone knows. Everyone knows the dishwashers, the drills, the battery packs, but not everyone knows that Bosch also has mobility aftermarket and wiper systems. I am a part of the wiper systems. I work in Tienen in Belgium, and we make wiper blades. We have four production sites for wiper blades all over the world. In Tienen, we are the only production site for rubber profiles. The rubber profiles we make for all the world, and we make around 120 million rubber profiles per year.

I'm part of the Mini Factory Rubber, which is actually… It's called Mini Factory because it's a factory on its own, and we have there four departments. We start with the product development, where they make the compounds and the profiles. Then you have the chemical process development, that's where I am a part of, where we do improvements on the processes or make new processes. Then we have two value streams, two production units in the batch process, they make the compound, and then in the next phase, they make the rubber profiles. Then the last department is our chemical laboratory.

In the application of JMP in the Six Sigma methodology, in the process development, we use actually the Six Sigma methodology, and we also use the enhance A3 methodology. The Six Sigma methodology we use when we need to focus on the statistical root cause analysis with the difficult problems and the enhanced A3 method we use when we need to focus on visualization.

The Six Sigma methodology contains of five different steps. The first step is the define step. This is a very important step, and it's a step which is often forgotten. There it's important that you define the problem and the target very specifically because everyone has quite often a different idea about what to achieve.

In the measure phase, we collect relevant data about the process and the problem, and in the analyze phase, we identify the cause and effect relationship between input and outputs. In the improve phase, we determine the optimum values for the process inputs. Then in the last phase, the control phase, we establish the standards to sustain improvements in the long run. Also, this phase is quite often forgotten, but if you do so and you don't do it properly, then your process will run away again.

You also have your tool box, and here you can see the DMAIC tool box. Each phase has specific tools and techniques associated with it, and this is an overview of the common tools. I will not go into detail, but just to know what you can do within the Six Sigma methodology. Elodie will take over now.

Hi, I'm Elodie. I'm a systems engineer at JMP, and I will share with you the marketplace we have with add-ins. As Marjoleine said, there is a procedure methodology which is very famous in Six Sigma, which is the define, measure, analyze, improve, and control methodology. We have an add-in in JMP that you can add in the software if you are interested in that methodology. Where can we find add-ins?

We launched last year a marketplace. This is the JMP marketplace. We have a search bar there where we can look for what we need. We have a Six Sigma add-in, which is called Modern Six Sigma Add-in. You can download it from there and then put it directly in JMP. It looks like this. It's a new menu which is added in your software, so here, with the five different steps. You have tool inside, and you also have available some webinars and help if you want to practice with this tool. Then I will turn things to Marjoleine.

I just spoke in the first phase that it was defined phase. The second phase is the measure phase. The measure phase is actually the phase where we quantify the current performance of the process and gather relevant data. Why is this important? Because often we have a gut feeling that it is a problem, and we are not always quite sure that this problem is really existing.

I have here the green boxes. The green boxes is actually the part of the Six Sigma methodology, which you can do in JMP. There are two things you can do there. There's, first of all, a measurement system analysis, where you can assess the accuracy and the precision of the measurement systems. The second part is actually of the measurement phase is the descriptive statistics, where you need to summarize the data using the mean, the median, the mode, whatever you need for your application.

I will now give a small example of an equivalence test of the mean between an old and a new measurement of about TOST, Two One-Sided Tests approach. Here you can see the data. We had 329 samples which we measured in two different ways. We had old application and a new application, and we wanted to see if there was a big difference between both methods.

You do that actually by analyze and then fit Y by X, and then you can go to the equivalence test. You need to give in what difference of the mean is practical important. In this case, it was 5 millimeters. What we could see if we do the TOST, normally, if it is equivalent, you need to be in the blue area in the forest plot.

In this case, we were not in the blue area, we were in the red area, and that was because the actual difference in the mean was 10 millimeters. Therefore, we needed to conclude actually that our old annual application measurement system data was not the same and that we needed to do some extra steps.

Also, the descriptive statistics, and the description of the current process, and the confirmation of the problem statement. I can show you one. This here was a zero series we made. We did that in different days, and we actually wanted to know if the measurement was indeed higher than we had previously during research.

If I look at the distribution here, we could see that we had a mean value of around 97 millimeters, where we had... In the previous sessions, we had 50 millimeters. Indeed, our gut feeling was right, we had a problem.

Then the third phase, that's the analyze phase, which is to identify the root causes of defects or performance issues. The first thing we always do is that we identify the root causes. We do that by using Ishikawa or 5 Why or a fishbone diagram, but that's always the base.

Then in the second phase, we will make a Pareto chart to identify the most significant factors in the data set, the three most important factors. With those three most important factors, we will then do the further analysis. Then Elodie will give some explanation about the Pareto chart.

I will speak about the Pareto chart. The objective of a Pareto chart is to analyze the frequency of the problem or the causes to understand which are the most frequent causes of problems. We can do that, thanks to the add-in. In general, the Pareto chart is included in the analyze phase. Here in our add-in, it's in the measure phase. Indeed, here in Modern Six Sigma menu. In the measure phase, we have the Pareto Chart. If you want to create a Pareto Chart, you can go through that menu.

I will present you an example. I have here a data table with different type of defects. I have 10 types of defects and the number of count for each defect. In that case, my Pareto chart can be like that. I have the cumulative percentages of my causes. It's ending at 100%, all the problems, and it's ordered by frequency from the highest to the lowest.

For example, we can see that the three main type of problems, are metal bridging, metal scratching, and metal defect. They are the main causes of problems. We identified them, and now we can work on them in the other steps. I will turn the slide and turn over to Marjoleine.

Another part of the analyze phase is the regression analysis, where we want to analyze the relationship between the different variables and also the hypothesis testing, where we test the assumptions and theories using statistical methods. Also from this one, I have an example.

First of all, we have here the same data from last time, the measurements from the zero series, and we had a feeling that there was a day-to-day variation in the measurement. The first thing what we did was actually to go to the Graph Builder and check our measurement versus date. In an histogram here, you can choose in the Graph Builder in your control panel, the ones you want, but we have taken here the box plots.

Already clear from the box plots, we see that there is a difference in the different dates. We wanted to be sure, so we did a hypothesis testing. The hypothesis testing, we did a one-way analysis on ANOVA, and you do that also in the fit Y by X, and then you can choose here means/ANOVA, and then analysis of means methods.

If you look here already in the data, you can see the diamonds, and the diamonds are not overlapping. That's already your biggest indication that there is indeed a day-to-day variance. We can also see it here in the values of the mean, the lower confidence and the upper 95% confidence intervals of the different sets are not overlapping, which is also clear that there is a day-to-day variance. You can also see it here in the analysis of means in the ANOVA graph, where you can see that they are not from the same behavior. They don't have the same behavior.

The fourth phase is the improvement phase, where we develop and implement the solutions to address the root causes and improve the process. Of course, first of all, you start with the brainstorming to get some ideas. This cannot be done in JMP, but then the second part will be a design of experiments. We often use the custom design of experiments, and we plan and conduct experiments to identify the factors influencing the outcomes.

We use quite often the custom design, and this has the reason actually that we want to do so, because with a custom design, you can also use hard to change factors. For us in production, this is very important. As soon as we have the information about the DOE, we go actually to the trial implementation of a part of the proposed solution on a reduced scale to check if our model is working correctly. I have here an explanation about actually a compounding study, which we do with the DOE.

In the past, before 2018, we did it in a one factor at a time method. It took us 20 weeks, 360 compounds, and we had a very high cost and quite often no success. After 2018, we started with an optimal design, an I-optimal design, response surface model. We can do that one in 2 weeks. It costs us 36 compounds, and our model can predict the optimal compound. We have 10% of labor consumption and lead-time in comparison to an OFAT approach, and we have more information than we had in the past.

I'm going to this model. Here you can see that we have a model with 10 output factors. These output factors are actually physical properties which can determine the functional behavior of this wiper. We have the five input factors or five input factors are actually our components of our compound. Then you can also see a blocking factor, and that has to do with the fact that we can only do nine tests in a day.

The model we use. Like I said, we use an I-optimal design, and we use a response surface model. In this case, the minimum amount of tests needed is, according to JMP, 27. We always say we take four more than 27, but in this case, we have anyway four test days, so 36 is a possibility for us. We will take the 36 like that in the default.

If we do then the analysis, this is the analyzed model we use. Let me move that one a little bit. You have here the effect summary. You can see that all P values are nicely below 0.05. This is an output. If we look at the output, you can see that the actual by-predicted plot is very small. It's a straight line like we expect, but also here you can see that the red bars are very, very small.

We can also see that in the summary of fit, we will always look at the R-square adjusted. For us, a model which is above 0.90 is a good model. In this case, we have a model which is 0.99, and this means that 99% of the variability can be explained by the model. If we analyze this model with 10 outputs and 5 inputs, for us, it's very important that we can have in the prediction profiler put in the desirability functions so that we can maximize the desirability and that we can see the optimum compound here.

For us, it's important that we know what the computer decides, but we always will have a look manually to see what are the interactions and so on. In this case, for example, we know that input 1 and input 2 have an interaction, and by moving the profiler, we can clearly see what is the influence of the different factors. You see here also in the outputs, you have very small confidence intervals here for certain outputs which have a very good R-square adjusted. We have also some other outputs which have a larger variability.

Then if you go to the last phase, so the control phase. Here in the control phase, we need to ensure that the improvements are sustained over time and that the process remains stable and controlled. A control plan is the first thing we need to check, if that's correct, adjust it if necessary, document the methods.

Then the second thing we always do is also statistical process control, where we use control charts to monitor this process stability. Within Bosch, we use a special program for it. Therefore, Elodie will also explain something about process control within JMP.

I will share how we can do control charts in the add-in. As we said, the control phase is very important in the last phase of this Six Sigma methodology. From the menu, it's available in the last phase control here. There are a lot of choices to control charts. You can either take control chart builder, which is a very customizable tool where you can do nearly any type of charts. Otherwise, if you want a specific type of chart, you can select one of the other options. I will present you an example on data and we will do a control chart.

Here, the data are a bit funny and fictitious. It's about pickles. We are measuring the acidity of the pickles. The measurements have been taken at different dates on different batches, and the objective is to know if the production is stable.

Here is the control chart from the control chart builder. It's divided into two subparts, the first one there, acid. It's the point which have been measured. They are all plotted. In red, the red lines, you can see the control limits. If a point is outside of the control limits, like this one, it's an alert that your process may not be stable, so you need to check what's happening.

You can also get some other alerts about stability or potential shifts in the mean or a potential problem coming. In advance, try to avoid that problem. Thanks to some other alerts. From JMP, you can customize your own alerts or use existing alerts. A lot of different types of alert are available.

Here we have two which are lighted. They are on, if I can say. This one, for example, it's number 5. You can have the meaning by hovering your mouse on the point. It's meaning that two over the three last points are beyond two Sigma after the mean. That may be a sign of a problem. You can be alerted and then check if it's a problem or not and avoid a potential problem.

Same, this is another type of alert, which is saying that you might have a shift on your process, so maybe it's not stable, or you can at least check. There is there another subpart which is also a control chart based on moving range. You can see if there is a too big or a suspicious range, if I can say, between two points.

It's also a way to check the variability of your process to be sure that your process remains stable because we did all the steps in one objective, and we want now our process to be stable and to stay stable. Now, it's the end of our presentation, so if you have questions.