The Design of Experiments or DOE platform helps us to design, organize and execute controlled experiments to improve products or procedures. At a basic level DOE utilizes a multi-factor approach that considers the interaction of those factors on a one or more responses. It answers a question like, what combination of system settings will result in an improvement to X in our final product? Like it now weighs less but has the same volume, it's better tasting or it's color is more appealing. Often this is a measurement we care about, in DOE this is called our Response, like an employee job performance rating, revenue dollars, amount of new customers, efficacy rate of medication, or low defect count in manufacturing. Virtually any product or process can benefit from DOE. Before you begin any DOE step, be sure to determine the goals or the responses of your experiment so you can accurately specify and generate a design consistent with your predictive model.
I’m sure there are some exceptions, however if you consider all of the factors and their interactions with each other you're certain bound to surface identify some improvement areas.
On Screen: (show Engineers, Scientists, Researchers or Product designers) The methods for determining how to increase productivity and improve quality are evolving. They have changed from costly and time-consuming trial-and-error searches to the powerful, elegant, and cost-effective statistical methods.
In JMP, there’s a DOE menu where we can choose then launch an experiment. We can pick from several classical text-book designs or customize our own design. Custom designs tend to be more flexible and cost effective when compared to classical designs. The Custom Design platform can construct many special design types, including, screening, mixture, response surface, split-split-plot and many more. In short, it lets you customize a design for your unique situation where you control the number of runs.
Not sure about adding this a a disclaimer?? (Sound engineering and process knowledge are critical to designing the experiment, running the experiments, and analyzing the results.)
On Screen: Designing experiments in JMP is centered around factors, responses, a model, and runs. JMP helps you determine if and how a factor affects a response.
If you have never used JMP to design an experiment the I recommended starting with the custom designer, the 1st option under the DOE menu.
You don't not need a data table open to launch and start an experiment in the DOE platform. JMP will be your guide you in creating a custom one.
After selecting Custom Designer we’ll first be asked the response and the factors or effects. The response is something you care about and you can measure. For example it could be an optimal amount of beverage in a packaging line at a soft drink bottler. Too much product, it spills over, not enough the customer will complain the didn’t get what they paid or worse yet, they'll never purchase it.
This is the same step Here we’ll also enter the factors or effects. Here’s the step where Wwe really need to think about or have thought about everything that we suspect that has variability in the output or procedure. We’ll enter those factors here.
Then we’ll add interactions, like how does the CO2 level interact with the fill rate. We can also disallow certain combinations. While learning, it’s best to run a few experiments to gain a full understanding and the best practices setting up your interactions.
In the next step, JMP then will determine the amount of runs we need to collect the results from. It says we need 12. We’ve determined and we have budget to for 15 total runs. Let's use 12 for this experiment and save 3 for our confirmatory experiment. DOE can save a considerable amount of money and time, however our desire is to have as few runs as possible because as each run can be expensive.
In this example, we have to use production time on our bottling line and employee’s hours.
We make this experiment into a data table and then we schedule our experiment with the plant manager schedule to conduct the runs in the 3rd shift minimizing production output.
On Screen - Illustrate that this isn’t your any ordinary data table though. Showing with arrows the pre-built model available in the tables panel, as well as the DOE dialog where we can return back and change up our design parameters.
After collecting our output we will then he’d back over to JMP and enter those findings in the data table.
Then we’ll run the included model.
We’ll then analyze the report from the model to understand what’s significant, or what’s not significant, looking 1st at the effect summary then interacting with Profiler. Profiler is powerful and intuitive tool to use, try setting Maximize Desraibility under the red triangle or dragging the dotted factor lines to see how the factors interact collectivelly on the response. You can also enter values in.
After fitting the model, it's time to test those optimal factor you discovered by conducting a confirmation experiment. So we have 3 runs left in our budget so we'll head back out to the bottling line adjust our factors, the machine settings, and verify if our fitted model in fact, works. You can have what appears to be a solid model however there may be some varaibility in something you have not discovered yet. So it's imperative to validate your findings. If you need to run a new expermeiment after learning there was something you did not account for you would head back to JMP with that new information. That's just part of the process.
If everything goes right, that we are statisfied with the results, we’ll have some happy customers with minimal product waste.
You can test drive the JMP Custom Designer with the tutorial thats built into JMP under the Help then Tutorials or open and explore DOE using the Coffee sample data.
You’ll then be ready to grab a your colleagues and apply DOE to your product or procedure to improve or optimize what you care about.
