Rockwell Collins has a Traveling Road Show that is available to schools free of cost - and it's the coolest thing you've ever seen. They have 3 engineers (actually more, but Ryan and I met with 3) with colorful personalities that present several different skits to teach kids how material engineering works. They demonstrate awesome chemical interactions, physical properties, how terminology and jargon was originated, and many more fun topics. We were told about their show, we didn't get to actually witness it, but the stories themselves were fun.
I've got to find someone at Jefferson who could use them coming in to match their curriculum. I asked them about a connection to Algebra 2, and they said they could find something that fit, but they didn't seem to have anything immediately in mind. It is mostly a science connection. I'll talk with John Hegarty about this first and go from there.
The Prediction Program for the time estimate of the printed circuit boards is coming along wonderfully. I have the "success" percentages up to a range of 60% to 82%! I think Jen will be pleased, but we'll find out tomorrow.
Ryan was able to get the Environmental Effects data to match the original data very well, so we are well on our way to being able to present that prediction tool as well.
We are on fire this week!
Friday, July 16, 2010
Wednesday, July 14, 2010
Day 23 (July 14, 2010)
When is frustration a good thing? When it makes my mind continue to think about a something when you you are not concentrating on it. My brain does that, and I've heard many of my students tell me their minds do the same. When you get to a point that the answer seems bleak, put it aside for awhile. Your brain will continue to work on the problem and when you come back to it, you may have come up with an answer or new approach.
Ryan has made terrific progress on recreating John's ECO count data. We should be able to present the process to John in the near future. Great job, Ryan!!!
My Touch Time and Cycle Time prediction tool is improving quite nicely today. I've gotten up to a 58% and 68% accuracy - much improved over what I was getting as results yesterday. Hip hip...
Ryan has made terrific progress on recreating John's ECO count data. We should be able to present the process to John in the near future. Great job, Ryan!!!
My Touch Time and Cycle Time prediction tool is improving quite nicely today. I've gotten up to a 58% and 68% accuracy - much improved over what I was getting as results yesterday. Hip hip...
Tuesday, July 13, 2010
Day 22 (July 13, 2010)
Frustration has set in... Ryan and I are hitting roadblocks on both of our projects.
We have been designing a prediction tool for the Touch Time and Cycle Time needed to complete a project. We thought by including the engineer and designer into the equation of the calculation, we could be more successful - and maybe we are more successful, but the final outcome cannot be considered a success. We are not very accurate. We consistently over estimate the touch and cycle times, but in many cases by a large amount.
In our other project of trying to predict the work load on the Environmental Effects lab, we are trying to recreate the data collected a year and a half ago. We are really struggling trying to make the data match. Arrrrrrgh!
We have been designing a prediction tool for the Touch Time and Cycle Time needed to complete a project. We thought by including the engineer and designer into the equation of the calculation, we could be more successful - and maybe we are more successful, but the final outcome cannot be considered a success. We are not very accurate. We consistently over estimate the touch and cycle times, but in many cases by a large amount.
In our other project of trying to predict the work load on the Environmental Effects lab, we are trying to recreate the data collected a year and a half ago. We are really struggling trying to make the data match. Arrrrrrgh!
Monday, July 12, 2010
Day 21 (July 12, 2010)
Reverse engineering is not an easy thing. Our new call is to recreate the data we used to predict the EMI annual hours, but we are not getting the data to match or even be in the vicinity of the supplied data. Today is going to be filled with figuring out how to create queries on an Access Database and pulling the data.
Ryan and I have agreed to skip trying to match the data and just use our newly found data and doing the same data analysis on this data that we had done on the supplied data.
Ryan and I have agreed to skip trying to match the data and just use our newly found data and doing the same data analysis on this data that we had done on the supplied data.
Friday, July 9, 2010
Day 20 (July 9, 2010)
I created a prototype prediction calculator today and it went much better than I expected. I made the calculator in Excel and used several formulas (like vlookups). Due to the use of these formulas, I was able to complete the job without having to create a macro. Ryan's contributions were useful. I think Ryan and I may need to spend a lot of our time over the next couple weeks expanding the calculator, but for now it seems to do a decent job at what it is expected to do so far.
I'm having trouble figuring out how to use the database Ryan and I gained access to yesterday. We are trying to figure out how to get the data John had pulled a year and a half ago. Which columns to select... Which columns to add together... Which columns to not pull... What do the abbreviations mean??? More time needs to be spent thinking about these things.
I'm having trouble figuring out how to use the database Ryan and I gained access to yesterday. We are trying to figure out how to get the data John had pulled a year and a half ago. Which columns to select... Which columns to add together... Which columns to not pull... What do the abbreviations mean??? More time needs to be spent thinking about these things.
Day 19 (July 8, 2010)
Ryan and I had a couple interesting experiences today. We attended a Learn at Lunch project about outdated Analog Computers and how they are still useful in some ways today. The point was do not forget how the analog worked because that can help make things easier in the more modern world.
We also got to see and experience a high tech flight simulator. The displays that are present on board actual aircraft along with the simulation displays. The program correlates the GPS mapping and the GPS system to determine the location of the plane so there can be an artificial display that mimics the real life position of the plane. This allows the pilot to fly even when real visibility is low.
The projects are moving ahead. I've started to write a program (macro) in Excel to give a prediction based on the entered criteria. I will then test to see how accurate the program is.
We also got to see and experience a high tech flight simulator. The displays that are present on board actual aircraft along with the simulation displays. The program correlates the GPS mapping and the GPS system to determine the location of the plane so there can be an artificial display that mimics the real life position of the plane. This allows the pilot to fly even when real visibility is low.
The projects are moving ahead. I've started to write a program (macro) in Excel to give a prediction based on the entered criteria. I will then test to see how accurate the program is.
Thursday, July 8, 2010
Day 18 (July 7, 2010)
"Interesting." "Encouraging." Those words were John's reaction to our presentation about predicting the work load the Engineering Effects department will be experiencing in the future. Our next steps are to figure our how to recreate the data John originally shared with us. That will be taking data from a huge database and summarizing it down into smaller categories. Now we are just waiting to gain access to the database.
One of my focuses today was to figure out how this job can be translated into the classroom. The obvious answer is to bring the data that we have been looking for the regression lines and give the students real-life data to our curriculum. The less obvious answer is to create some type of activity can be used to simulate the data which can then be used to practice analyzing "less than obvious" data.
My initial thought was to assign each student a certain number of dice and a criteria they must role to "finish" the job. They must count the number rolls needed to finish the job. Students will be randomly assigned how many dice to roll (rolling 1 die, 2 dice or 3 dice), their "ability" to complete the job, and how many jobs they must complete. Their ability will be based on having different probabilities of completing the job. For example, a 1 die rolling student must roll an even to complete the job, while another student rolling a single die must roll a 6 to complete the job. The "even" student is simulated to have more ability that the "6" student.
Other ideas Ryan brainstormed with me was using real Major league data and determine how many at bats did it take for each player to get their first hit (or first 10 hits or the like) and compare those results to their batting averages. Another was to have a list of words and give the student different rules to follow. i.e count the number of words that alphabetically come before M and how many after vs. determine if the word is a commonly used word or not. After them hearing the list and following their rule, have them try to recreate the list. One rule will typically be much more accurate than the other.
One of my focuses today was to figure out how this job can be translated into the classroom. The obvious answer is to bring the data that we have been looking for the regression lines and give the students real-life data to our curriculum. The less obvious answer is to create some type of activity can be used to simulate the data which can then be used to practice analyzing "less than obvious" data.
My initial thought was to assign each student a certain number of dice and a criteria they must role to "finish" the job. They must count the number rolls needed to finish the job. Students will be randomly assigned how many dice to roll (rolling 1 die, 2 dice or 3 dice), their "ability" to complete the job, and how many jobs they must complete. Their ability will be based on having different probabilities of completing the job. For example, a 1 die rolling student must roll an even to complete the job, while another student rolling a single die must roll a 6 to complete the job. The "even" student is simulated to have more ability that the "6" student.
Other ideas Ryan brainstormed with me was using real Major league data and determine how many at bats did it take for each player to get their first hit (or first 10 hits or the like) and compare those results to their batting averages. Another was to have a list of words and give the student different rules to follow. i.e count the number of words that alphabetically come before M and how many after vs. determine if the word is a commonly used word or not. After them hearing the list and following their rule, have them try to recreate the list. One rule will typically be much more accurate than the other.
Friday, July 2, 2010
Day 15 (July 2, 2010)
Today marks the half way mark of my experience here. I am pleased with my progress so far as far as understanding the work involved, but trying to make connections to the classroom have not been overly obvious. I will take more time in the next couple weeks concentrating on that aspect.
Today, Ryan and I met with Jenny Becker, the Community Relations person who originally introduced us to the program. Her job is to bring the community and Rockwell Collins personnel together. She showed us a couple fabulous programs. One is a program for high elementary to middle school aged students (called FIRST LEGO League) and the other is for high school level students (FIRST Tech Challenge, aka FTC). The FTC is actually a class at Jefferson this coming year. I'll make a note to look in on the program. I need to look over my curricula a little closer and see if there is a good fit for a Rockwell Collins employee to visit my my class(es).
On the analysis front, I have put together a presentation to show John, the head of the Environmental Effects department. I think I may have a tool for him to use to predict the number of hours to plan for in his area for the fiscal year 2011. He technically has already made the decision (he was forced to early last week), but I will be curious if my conclusion matches his estimation.
Rockwell Collins is closed for Monday and Tuesday and are considered paid holidays. My next blog entry will be on Wednesday, July 7. See you then!
Today, Ryan and I met with Jenny Becker, the Community Relations person who originally introduced us to the program. Her job is to bring the community and Rockwell Collins personnel together. She showed us a couple fabulous programs. One is a program for high elementary to middle school aged students (called FIRST LEGO League) and the other is for high school level students (FIRST Tech Challenge, aka FTC). The FTC is actually a class at Jefferson this coming year. I'll make a note to look in on the program. I need to look over my curricula a little closer and see if there is a good fit for a Rockwell Collins employee to visit my my class(es).
On the analysis front, I have put together a presentation to show John, the head of the Environmental Effects department. I think I may have a tool for him to use to predict the number of hours to plan for in his area for the fiscal year 2011. He technically has already made the decision (he was forced to early last week), but I will be curious if my conclusion matches his estimation.
Rockwell Collins is closed for Monday and Tuesday and are considered paid holidays. My next blog entry will be on Wednesday, July 7. See you then!
Thursday, July 1, 2010
Day 14 (July 1, 2010)
Another day in the cube - but good things happening! I decided to include some data considered "old" in order to have more data significance and found some interesting things. Full disclosure: "old" data simply means the predictive data was counted differently (and probably not as accurately as the newer method).
Yesterday I suggested the project completion time may be 3 1/2 years (actually 47 months). After including the old data, that number has changed to 25 months (just over 2 years). I am able to see a definite trend between the predictive data from 25 months ago and the Environment Effects hours now relatively. I will present the findings to John tomorrow.
I spent most of the day creating supporting graphs and making sure they were labeled well...
One of the things working with Ryan has helped me understand is the need for labels on graphs. I call him over when I think I have something significant and his first question is always, "What am I looking at?" The graphs make sense to me, but he has no clue. the graphs look pretty and significant, but knowing what it actually says is impossible without proper labeling. I have always emphasized that that in my classroom, but I see why it is important now. It's a communication tool, not just an analysis tool for me.
Yesterday I suggested the project completion time may be 3 1/2 years (actually 47 months). After including the old data, that number has changed to 25 months (just over 2 years). I am able to see a definite trend between the predictive data from 25 months ago and the Environment Effects hours now relatively. I will present the findings to John tomorrow.
I spent most of the day creating supporting graphs and making sure they were labeled well...
One of the things working with Ryan has helped me understand is the need for labels on graphs. I call him over when I think I have something significant and his first question is always, "What am I looking at?" The graphs make sense to me, but he has no clue. the graphs look pretty and significant, but knowing what it actually says is impossible without proper labeling. I have always emphasized that that in my classroom, but I see why it is important now. It's a communication tool, not just an analysis tool for me.
Wednesday, June 30, 2010
Day 13 (June 30, 2010)
Is a 0.92 correlation good? Why, it is great! I think I may be onto something on predicting the Environmental Effects department's annual hours. I have a data category called
Engineering Change Order (ECO) that seems to predict the actual hours about 3 1/2 years in the future. The ECO are simply the new projects started. Evidently it must take around 3 1/2 years to have the project ready for testing. The data is not 1-to-1, meaning the ECO numbers do not equal the EEE hours, but the data generally goes in the same direction.
I also have a prediction tool in place for the Printed Circuit Boards (PCB) cycle time and touch times. My predictors are not excellent at being exactly correct, but generally, the tool predicts correctly more often than the current tool (about 35% to 10%), but since an large overestimate is considered not as serious as an large underestimate, I got the number of catastrophic failures down to 4 to 10%. That means I am overestimating around 55 to 60% of the time. This is not hitting the final goal, so more work ahead for Ryan and me.
Ryan is currently cleaning the historical engineering and designer data so the entries by the same person are actually the same. We want to estimate the experience of each one and the only way to do this is to clean-up the data. He has worked diligently on this. The data is 50000+ lines long. I think the data will be great to have and see if experience correlates with the Schnick Rating I referred to yesterday. we are speculating that it will.
Engineering Change Order (ECO) that seems to predict the actual hours about 3 1/2 years in the future. The ECO are simply the new projects started. Evidently it must take around 3 1/2 years to have the project ready for testing. The data is not 1-to-1, meaning the ECO numbers do not equal the EEE hours, but the data generally goes in the same direction.
I also have a prediction tool in place for the Printed Circuit Boards (PCB) cycle time and touch times. My predictors are not excellent at being exactly correct, but generally, the tool predicts correctly more often than the current tool (about 35% to 10%), but since an large overestimate is considered not as serious as an large underestimate, I got the number of catastrophic failures down to 4 to 10%. That means I am overestimating around 55 to 60% of the time. This is not hitting the final goal, so more work ahead for Ryan and me.
Ryan is currently cleaning the historical engineering and designer data so the entries by the same person are actually the same. We want to estimate the experience of each one and the only way to do this is to clean-up the data. He has worked diligently on this. The data is 50000+ lines long. I think the data will be great to have and see if experience correlates with the Schnick Rating I referred to yesterday. we are speculating that it will.
Tuesday, June 29, 2010
Day 12 (June 29, 2010)
I have created something that needs a little modification, but gives a better estimation (with an emphasis to overestimate) that the currently used tool. I have created a predictive equation for the amount of touch time it will take to complete a circuit board project. Then the number is adjusted (this is the part that needs modification) based on the engineer and the designer. If both are highly rated, the time estimate can be decreased, perhaps significantly. If both are rated low, the estimate must be increased significantly. The predictive number is based on an average engineer and an average designer.
I may be onto a break through with the Environmental Effects Engineering lab time prediction. I found a unique high correlation with the total hours and another predictive variable. The predictive variable had to be time adjusted, so I spent most of the afternoon setting up a spreadsheet to make that time adjustment. It seems to be working well now, but I need a little guidance from John, the head of the EEE department. If he responds the way I am hoping, I will present my findings to him. The problem with the predictive variable is that I can only be comfortable predicting to within about 40% accuracy. This may or may not be acceptable. I hope to find out tomorrow.
Ryan and I are working well as a team. We are both working in different directions, but keeping each other aware of what we are doing. I have given Ryan some Excel pointers and he had given me some excellent ideas about analyzing the data. He's also set me straight on a few of my stray ideas. When I brainstorm, I get a little carried away and misinterpret the data sometimes. He's reeled me in more than once when this happens.
I may be onto a break through with the Environmental Effects Engineering lab time prediction. I found a unique high correlation with the total hours and another predictive variable. The predictive variable had to be time adjusted, so I spent most of the afternoon setting up a spreadsheet to make that time adjustment. It seems to be working well now, but I need a little guidance from John, the head of the EEE department. If he responds the way I am hoping, I will present my findings to him. The problem with the predictive variable is that I can only be comfortable predicting to within about 40% accuracy. This may or may not be acceptable. I hope to find out tomorrow.
Ryan and I are working well as a team. We are both working in different directions, but keeping each other aware of what we are doing. I have given Ryan some Excel pointers and he had given me some excellent ideas about analyzing the data. He's also set me straight on a few of my stray ideas. When I brainstorm, I get a little carried away and misinterpret the data sometimes. He's reeled me in more than once when this happens.
Monday, June 28, 2010
Day 11 (June 28, 2010)
This was a data analysis day. My personal approach was to set up a "Schnick Rating" (a made up name using my nickname) on the predictability of the designed and the engineer. Basically, if they are consistently taking less time than expected complete a project, they get a high rating. The expected value is based on the Weka producted formula. If they are consistently well above the expected time, they get a low rating. If they are inconsistent or consistently hitting the expected time, they get a middle-of-the-road rating.
At the end of the day here, I've learned I've been using the wrong data to identify the engineer. I'll correct this mistake in the morning. This could be frustrating, but I realize this is still a work in progress and I'm learning something new everyday.
At the end of the day here, I've learned I've been using the wrong data to identify the engineer. I'll correct this mistake in the morning. This could be frustrating, but I realize this is still a work in progress and I'm learning something new everyday.
Friday, June 25, 2010
Day 10 (June 25, 2010)
Today was the first installment of the chemistry portion of how PCBs (printed circuit boards) are analyzed. The other day, Ryan and I where introduced to the EEE (Environments effects Engineering) department and the huge machines and rooms used there, but the AE (Application Engineering) department does some similar things, just not on such a large scale. The AE department seems to be more focused on long term effects on the components that are used to produce the boards, not the actual finished products.
Where the EEE simply says (at least this is my interpretation of what they do) fail or pass on a complete PCBs, the AE tries to determine the best components to use for a particular project. Based on temperature extremes and changes (including rate of change) or other potentially product disrupting possibilities, they can give suggestions on which components to use to have a higher potential for success.
Another function of AE is to determine the cause of failure. My understanding is that EEE does not necessarily do that, but rather leaves the fixing up to the designing engineer. Our contact in AE, Roy, had many great stories about the problem solving they have to do. There are different potential causes for failure: bad product, damaged product, product contamination, exceeding the product's safe parameters, and other product related reasons. However, Roy talked about several times where the tests themselves were flawed due some unusual circumstances. One example was that the air intake for a refrigeration unit was compromised because the warmed up by a worker taking off there shoes and resting their feet over the intake register. This simple little issue caused the air taken into the machine to raise in temperature enough that the machine could not get the air cold enough for the test. It only took a 20-hour marathon work period for Roy to figure that one out. He said he had the epiphany as to what was causing the problem at 3AM. I'm sure you don't read about that kind of problem solving in those engineer-training text books.
Where the EEE simply says (at least this is my interpretation of what they do) fail or pass on a complete PCBs, the AE tries to determine the best components to use for a particular project. Based on temperature extremes and changes (including rate of change) or other potentially product disrupting possibilities, they can give suggestions on which components to use to have a higher potential for success.
Another function of AE is to determine the cause of failure. My understanding is that EEE does not necessarily do that, but rather leaves the fixing up to the designing engineer. Our contact in AE, Roy, had many great stories about the problem solving they have to do. There are different potential causes for failure: bad product, damaged product, product contamination, exceeding the product's safe parameters, and other product related reasons. However, Roy talked about several times where the tests themselves were flawed due some unusual circumstances. One example was that the air intake for a refrigeration unit was compromised because the warmed up by a worker taking off there shoes and resting their feet over the intake register. This simple little issue caused the air taken into the machine to raise in temperature enough that the machine could not get the air cold enough for the test. It only took a 20-hour marathon work period for Roy to figure that one out. He said he had the epiphany as to what was causing the problem at 3AM. I'm sure you don't read about that kind of problem solving in those engineer-training text books.
Thursday, June 24, 2010
Day 9 (June 24, 2010)
Today was mostly a "put-your-nose-to-the-grind-stone" day. Getting into the Weka data mining software more thoroughly, I was able to get a correlation of .93 between my prediction on Touch Time and actual Touch Time on the New Digital Layouts. I'm digging into it a little more before I call it a success, but it looks very promising.
Ryan and I met with Ted Neal today to discuss how things are going. Ryan and I concurred that things are going excellently from our perspective. We are looking at some additional activities outside our job scopes, but they will be very hands on and educational.
It is great to know we are doing exactly the things the description of the position stated on the application. I had not reviewed the document since filling out the application, so seeing that the job description matches perfectly is pretty cool.
Ryan and I met with Ted Neal today to discuss how things are going. Ryan and I concurred that things are going excellently from our perspective. We are looking at some additional activities outside our job scopes, but they will be very hands on and educational.
It is great to know we are doing exactly the things the description of the position stated on the application. I had not reviewed the document since filling out the application, so seeing that the job description matches perfectly is pretty cool.
Day 8 (June 23, 2010)
Ryan and I met with John Thoreson today to get a better insight of our second project scope and goal. To be honest with you, Ryan and I are still a little confused on the expectations, but we agree that he wants us to come up with a new way to predict his department's (EEE) work load either on a monthly, quarterly, or even annually basis. We also got a tour through the facility with the machines and rooms that the testing facility rooms.
The EEE (Environmental Effects Engineering) has nothing to do with being "green". Rather, this is the testing department on products before they are produced in bulk and sent out to the customers. The tests are categorized into 3 groups: Climatics, Dynamics, and Electro-Magnetic. They test these products on climatic effects like extreme temperatures and temperature change, extreme air pressures and air-pressure changes, high humidity and high corrosive situations. The dynamic tests include G-force and high impact effects on the components. The electro-magnetic tests deal with high energy blasts and continuous high-energy environmental effects on the product as well as tests on the energy and "noise" produced by the product itself so it has minimal negative influence on the other components around them.
The EEE has a very interesting business concept. The goal is to have exactly zero profit. That means the supply and demand effects go against everything I've been taught about business concept. When EEE has a busy year, they are able to drop the price of the testing (aka high demand actually produces lower prices). Conversely, when demand is low, EEE may have to raise their prices to prevent a negative bottom line. They try to avoid reducing to the labor force at all costs because they are so valuable and it is difficult to train them in this department. It is very specialized.
Ryan and I each downloaded some data mining software which is on the internet as freeware. It has proved interesting. The program will take given information and look for correlations that may not be easy to see with the naked eye. We are just getting started learning how the program works, so details will have to wait for another day.
A statement Ryan asked me at the end of the day has got me perplexed a bit. He said the program found that as the number of parts increases the time measurment actually decreases. This goes against my early instincts on the Printed Curcuit Board production data. We'll talk about it tomorrow, I'm sure.
The EEE (Environmental Effects Engineering) has nothing to do with being "green". Rather, this is the testing department on products before they are produced in bulk and sent out to the customers. The tests are categorized into 3 groups: Climatics, Dynamics, and Electro-Magnetic. They test these products on climatic effects like extreme temperatures and temperature change, extreme air pressures and air-pressure changes, high humidity and high corrosive situations. The dynamic tests include G-force and high impact effects on the components. The electro-magnetic tests deal with high energy blasts and continuous high-energy environmental effects on the product as well as tests on the energy and "noise" produced by the product itself so it has minimal negative influence on the other components around them.
The EEE has a very interesting business concept. The goal is to have exactly zero profit. That means the supply and demand effects go against everything I've been taught about business concept. When EEE has a busy year, they are able to drop the price of the testing (aka high demand actually produces lower prices). Conversely, when demand is low, EEE may have to raise their prices to prevent a negative bottom line. They try to avoid reducing to the labor force at all costs because they are so valuable and it is difficult to train them in this department. It is very specialized.
Ryan and I each downloaded some data mining software which is on the internet as freeware. It has proved interesting. The program will take given information and look for correlations that may not be easy to see with the naked eye. We are just getting started learning how the program works, so details will have to wait for another day.
A statement Ryan asked me at the end of the day has got me perplexed a bit. He said the program found that as the number of parts increases the time measurment actually decreases. This goes against my early instincts on the Printed Curcuit Board production data. We'll talk about it tomorrow, I'm sure.
Tuesday, June 22, 2010
Day 7 (June 22, 2010)
Today I attended a class and tour on the manufacturing of the curcuit boards we are trying to estimate the cycle time and the touch time needed to produce. A couple things that came up during the tour illustrated as to why we have such an inconsistency on cycle time. The inventory for the boards can be held up when materials are not available - and this happens daily according to the guide. The manufacturing end is trying to predict how much demand there will be for each inventory item, but they are not always accurate.
The biggest problem with the inventory issue is that if they need to order new material it can take 2 to 6 weeks to receive it. Basically, Rockwell Collins is at the mercy of the supplier.
Ryan and I are continuing to look for better prediction methods on cycle time and touch time. I have made some progress using the logarithm of the number of parts multiplied by the number of board layers, but it is not great. More work ahead.
The biggest problem with the inventory issue is that if they need to order new material it can take 2 to 6 weeks to receive it. Basically, Rockwell Collins is at the mercy of the supplier.
Ryan and I are continuing to look for better prediction methods on cycle time and touch time. I have made some progress using the logarithm of the number of parts multiplied by the number of board layers, but it is not great. More work ahead.
Monday, June 21, 2010
Day 6 (June 21, 2010)
My partner has arrived! Ryan Hickerson from Linn-Mar has joined me today. He had orientation this morning and I helped a little in acclimating him to our area and building. He is going to be a great person to work with on this project.
Ryan and I were introduced to a second project we will be analyzing and probably trying to find a better method for accomplishing the goal. The goal is to predict the work load of the testing department, Environmental Effects Engineering (EEE). They have struggled in the past because the workload varies dramatically week to week and even quarter to quarter. We will be working directly with John Thoreson for a couple weeks starting next week.
This morning I did some additional analysis on the Touch Time and Cycle Time prediction of the Circuit Board development and may be on the verge of a breakthrough. Time will tell. First of all, I looked at the data broken into smaller groups (grouping the data based on similar Touch Times and Cycle Times independently) and looking for correlations with the data items used to make those predictions. What I found was very interesting. The correlations to measurements of time actually got worse (closer to zero) generally. This is very similar to fractal mathematics. As you zoom in, the edges actually get more crooked, not less. I wish I had more expertise in fractals. Perhaps this could be an avenue we want to pursuit if we cannot find another alternative.
Interestingly, the correlations of these cycle time fragments actually improved, quite dramatically in many cases, the correlations toward touch time. This initially gave me cause for celebration - I thought I was onto something. But by combining some of these small groupings together made the touch time correlations return closer to the entire population correlations. So the only way this works well is if we can pinpoint the cycle time initially, which is not something we are capable of at this time, then we could estimate the touch time more accurately.
I then started breaking down the data based on similar design and activity types and see if there is any consistency there. I could find nothing initially, so I looked at combining some data together. This showed some signs of success, but not enough. I then started transforming the
data mathematically (specifically logging one of the data). That is showing some signs of a better predictor. I ran out of time to develop it further today. Perhaps I can work with Ryan tomorrow on this pursuit.
Ryan and I were introduced to a second project we will be analyzing and probably trying to find a better method for accomplishing the goal. The goal is to predict the work load of the testing department, Environmental Effects Engineering (EEE). They have struggled in the past because the workload varies dramatically week to week and even quarter to quarter. We will be working directly with John Thoreson for a couple weeks starting next week.
This morning I did some additional analysis on the Touch Time and Cycle Time prediction of the Circuit Board development and may be on the verge of a breakthrough. Time will tell. First of all, I looked at the data broken into smaller groups (grouping the data based on similar Touch Times and Cycle Times independently) and looking for correlations with the data items used to make those predictions. What I found was very interesting. The correlations to measurements of time actually got worse (closer to zero) generally. This is very similar to fractal mathematics. As you zoom in, the edges actually get more crooked, not less. I wish I had more expertise in fractals. Perhaps this could be an avenue we want to pursuit if we cannot find another alternative.
Interestingly, the correlations of these cycle time fragments actually improved, quite dramatically in many cases, the correlations toward touch time. This initially gave me cause for celebration - I thought I was onto something. But by combining some of these small groupings together made the touch time correlations return closer to the entire population correlations. So the only way this works well is if we can pinpoint the cycle time initially, which is not something we are capable of at this time, then we could estimate the touch time more accurately.
I then started breaking down the data based on similar design and activity types and see if there is any consistency there. I could find nothing initially, so I looked at combining some data together. This showed some signs of success, but not enough. I then started transforming the
data mathematically (specifically logging one of the data). That is showing some signs of a better predictor. I ran out of time to develop it further today. Perhaps I can work with Ryan tomorrow on this pursuit.
Friday, June 18, 2010
Day 5 (June 18, 2010)
This week I have been spending time analyzing the amount of Touch Time and the Cycle Time for the production of engineered circuit boards. I thought of something overnight that I had not taken into consideration the first time I had analyzed the "actual" data. There are different stages of the design (for example, new or a relayout) and the type of board that is being created (for example, digital or analog).
This morning, I reran my randomly selected actuals and record the outcomes. I did find that my results did change, and slightly for the better as far as the tool was a little better at predicting with the design stage and type of board information is entered. It was not, however, significantly better and still rather poor at predicting.
I put together the data into a formal statistical test. Using the method we used in my AP Stats class, I drew the conclusion that the tool was at best 15% accurate in predicting the Touch Time and 18% accurate in predicting the Cycle Time. This is not good and is not what my department wants.
I was able to present this information to some of the people this tool effects. Their instincts had told them the tool was not good, but this was verification of that instinct.
I then showed them why the tool is not good. The reason is not the tool. The reason for the inaccuracy is the inconsistency of the actual data. I showed them several of specific instances of these inconsistencies. So I drew the conclusion that the current information entries into the tool do not correlate well with the two measures of time. I found some better correlated data, but I'm not convinced even having those data predicted would give a better result.
Overall, I think they liked my presentation of the data. There were a few additional ideas to come out of the meeting: 1) to look for a correlation between a data value and the ratio of a couple others and 2) to compare if jobs that take a long cycle time boards are easier or more difficult to predict than the shorter cycle time boards.
This morning, I reran my randomly selected actuals and record the outcomes. I did find that my results did change, and slightly for the better as far as the tool was a little better at predicting with the design stage and type of board information is entered. It was not, however, significantly better and still rather poor at predicting.
I put together the data into a formal statistical test. Using the method we used in my AP Stats class, I drew the conclusion that the tool was at best 15% accurate in predicting the Touch Time and 18% accurate in predicting the Cycle Time. This is not good and is not what my department wants.
I was able to present this information to some of the people this tool effects. Their instincts had told them the tool was not good, but this was verification of that instinct.
I then showed them why the tool is not good. The reason is not the tool. The reason for the inaccuracy is the inconsistency of the actual data. I showed them several of specific instances of these inconsistencies. So I drew the conclusion that the current information entries into the tool do not correlate well with the two measures of time. I found some better correlated data, but I'm not convinced even having those data predicted would give a better result.
Overall, I think they liked my presentation of the data. There were a few additional ideas to come out of the meeting: 1) to look for a correlation between a data value and the ratio of a couple others and 2) to compare if jobs that take a long cycle time boards are easier or more difficult to predict than the shorter cycle time boards.
Thursday, June 17, 2010
Day 4 (June 17, 2010)
As a statistician you are always looking for patterns and making things predictable. I have been beating my head against the wall for the past 3 days trying to see if the current time prediction process is working and my conclusion is that the data is so haphazard given the known inputs that making a reasonable prediction is unrealistic. I feel a little better about things at this point because I was told that is an okay conclusion. That is the conclusion my coordinator was expecting actually. Whew, what a relief! I am going to turn my attention to trying to create a more reliable predictor.
Some terminology that is part of this industry can easily be tied to my AP Stat class. One of the significant features of statistics is to look at the normal curve. Part of the normal curve is the 68-95-99.7 concept. That is, 68% of the data is within 1 standard deviation (sometimes referred to as within 1 sigma - the Greek symbol commonly used to represent standard deviation), 95% is with 2 standard deviations (2 sigmas), and 99.7% is within 3 standard deviations (3 sigmas). The industry term that applies here is "6 sigma". 99.9999997% of all the data is within 6 standard deviations from the mean. In other words, it is nearly impossible for the data not to fit within that parameter. It makes me feel pretty good about getting into the airplane the next time I need to travel because Rockwell Collins supplies circuit boards for those planes.
I ran into a former student's father today and he wanted to show me what he is working on. He is testing a flight simulator. He has to make sure the simulator works for every possible combination of settings. He has already spent 4 months going through the documentation and was only about 3/4 of the way through the about 400 page binder. He has several tags on the pages indication he needs someone else to look at portion as he was not able to verify that it worked. It sounds like very tedious work, but he does get to play on a flight simulator all day.
I attended a presentation on Circuit Board production today. It was specifically about the industry regulation-led change on the type of solder used to hold components in place on the circuit boards. There is a shift from tin-lead solder to lead-free solder. The new type solder itself apparently works almost as well as the previous with the exception that the lead-free solder tends to grow "tin-whiskers". These tin whiskers cause many electrical issues if they touch other metal portions of the board or casing. Another problem with the lead-free solder is the solder gun must be set at a higher temperature to melt it so the manufacturer can get the solder in place. This can cause overheating problems for the components or even the board if they are not adjusted to be able to withstand the higher temperature.
Some terminology that is part of this industry can easily be tied to my AP Stat class. One of the significant features of statistics is to look at the normal curve. Part of the normal curve is the 68-95-99.7 concept. That is, 68% of the data is within 1 standard deviation (sometimes referred to as within 1 sigma - the Greek symbol commonly used to represent standard deviation), 95% is with 2 standard deviations (2 sigmas), and 99.7% is within 3 standard deviations (3 sigmas). The industry term that applies here is "6 sigma". 99.9999997% of all the data is within 6 standard deviations from the mean. In other words, it is nearly impossible for the data not to fit within that parameter. It makes me feel pretty good about getting into the airplane the next time I need to travel because Rockwell Collins supplies circuit boards for those planes.
I ran into a former student's father today and he wanted to show me what he is working on. He is testing a flight simulator. He has to make sure the simulator works for every possible combination of settings. He has already spent 4 months going through the documentation and was only about 3/4 of the way through the about 400 page binder. He has several tags on the pages indication he needs someone else to look at portion as he was not able to verify that it worked. It sounds like very tedious work, but he does get to play on a flight simulator all day.
I attended a presentation on Circuit Board production today. It was specifically about the industry regulation-led change on the type of solder used to hold components in place on the circuit boards. There is a shift from tin-lead solder to lead-free solder. The new type solder itself apparently works almost as well as the previous with the exception that the lead-free solder tends to grow "tin-whiskers". These tin whiskers cause many electrical issues if they touch other metal portions of the board or casing. Another problem with the lead-free solder is the solder gun must be set at a higher temperature to melt it so the manufacturer can get the solder in place. This can cause overheating problems for the components or even the board if they are not adjusted to be able to withstand the higher temperature.
Wednesday, June 16, 2010
Day 3 (June 16, 2010)
Today was a change of gears. As my purpose here at Rockwell Collins is being molded, I understand the goal of my project much better. I am to be testing the current estimation tool and found out the tool is not a reliable predictor given its current status. I have started getting into the programming code and data files to determine what exactly it is looking at to make those predictions. The process has been quite the challenge so far.
I have been a little surprised at what I think is going on with the regression problem. It seems to only be regressing over a single variable after filtering the data over 6 variables. I plan to investigate if a multi-variable regression would be more accurate, but until I can narrow down what the program is actually looking at, I cannot make that suggestion.
I was exposed to computer code today (something I am not very proficient in) and was able to locate the data filtering portion and the regression portion. I just started processing through the code with an example to see if I can replicate what the program is doing. No success yet.
Chuck Basset is a great colleague. He has worked hard to get me the data and programs I need.
I have been a little surprised at what I think is going on with the regression problem. It seems to only be regressing over a single variable after filtering the data over 6 variables. I plan to investigate if a multi-variable regression would be more accurate, but until I can narrow down what the program is actually looking at, I cannot make that suggestion.
I was exposed to computer code today (something I am not very proficient in) and was able to locate the data filtering portion and the regression portion. I just started processing through the code with an example to see if I can replicate what the program is doing. No success yet.
Chuck Basset is a great colleague. He has worked hard to get me the data and programs I need.
Day 2 (June 15, 2010)
Today I learned a tremendous amount by reading several parts of books and papers. I am starting to realize Printed Circuit Boards are the focus of my group (so I now know what PCB stands for). I read a few chapters from a booklet titled Printed Circuit Board Basics just to get an idea of the type of products my group produces. It was valuable, but not as valuable as the time Chuck Bassett spent with me showing me some of the terminology and what is actually was on a circuit board.
I reread Chapter 14 out of my AP Stat book to get me a better idea of Multivariate Regression, which is what I think will be the process I will be shooting for to complete my first product. But that is only a guess at this point.
I ran regression calculations on a large number of variable versus the two different measures of time used in manufacturing (Cycle Time and Touch Time). I found the process they have been using does not correlate as well as some other possibilities. My job tomorrow and the next day is to come up with a way to better predict these two measures of time given the values of certain variables.
I met with Michael Molacek to get an idea of what he wants from this product. He says I have free reign, just try to come up with a better prediction tool. That is my goal. My biggest problem is the data is partial or sporadic. My thoughts are to remove the partial data and use only the data which is complete to make my predictions. I keep you posted.
I reread Chapter 14 out of my AP Stat book to get me a better idea of Multivariate Regression, which is what I think will be the process I will be shooting for to complete my first product. But that is only a guess at this point.
I ran regression calculations on a large number of variable versus the two different measures of time used in manufacturing (Cycle Time and Touch Time). I found the process they have been using does not correlate as well as some other possibilities. My job tomorrow and the next day is to come up with a way to better predict these two measures of time given the values of certain variables.
I met with Michael Molacek to get an idea of what he wants from this product. He says I have free reign, just try to come up with a better prediction tool. That is my goal. My biggest problem is the data is partial or sporadic. My thoughts are to remove the partial data and use only the data which is complete to make my predictions. I keep you posted.
Day 1 (June 14, 2010)
Today I was introduced to the Rockwell Collins culture through the regular new employee orientation process. I was in a room with about 30 other new employees. Some of these new employees were internship and co-op employees with other companies.
I had some great conversation with newly hired engineers (Steve S.), one of which wants to discuss my externship opportunity with me. His wife is in the educational outreach for her company (I do not recall the name at this time) and could potentially be interested in a connection with this program. I’ll make note of any future conversation with him about this program. He suggested lunch in a couple weeks, so we’ll hopefully he follows through.
My orientation to the Rockwell Collins culture has been easy so far. The emphasis on ethics during the orientation was similar to the policy and expectation at my previous employer, AEGON. It would be wonderful to be able to have a similar code of ethics emphasis and training (perhaps we did in college, I just do not recall) for teachers. If we could implement the code down to the student level, that would be all the better.
I spent the afternoon watching online Ethics Training videos. They bring up some interesting scenarios that are not always easy to determine to immediate proper course of action. It was educational, but for the most part will not pertain to me in terms of customer contact.
The building I work in is huge and easy to get lost in. My immediate supervisor (Jen Waskow) gave me a tour of the building and even treated me to lunch in the cafeteria. She introduced me to her immediate Reports (people I am anticipating working with at least occasionally) and one of her supervisors (Michael Molacek). I will be working with Michael on one of the three statistical analyses projects they will have for Ryan and me. Note: Ryan’s first day has been delayed 1 week, so hopefully I’ll be able to get him up to speed when he gets here.
One of the projects Jen explained to me sounded very simple if all the data is corroborated easily, but I’m getting some indications that just orientating the data into a usable form may be the challenge.
Another project Jen explained to me (Michael’s project) it seems will be using Multi-vitiate data. That is just short of an AP Statistics curriculum item, so I’ll go back and read that chapter in the text book tomorrow (Chapter 14 – for some reason I remember the chapter).
Just before the end of the day, Jen suggested I Google “Business Process Metrics” to get me up to speed on some of the topics we will be looking at over the next few weeks.
I had some great conversation with newly hired engineers (Steve S.), one of which wants to discuss my externship opportunity with me. His wife is in the educational outreach for her company (I do not recall the name at this time) and could potentially be interested in a connection with this program. I’ll make note of any future conversation with him about this program. He suggested lunch in a couple weeks, so we’ll hopefully he follows through.
My orientation to the Rockwell Collins culture has been easy so far. The emphasis on ethics during the orientation was similar to the policy and expectation at my previous employer, AEGON. It would be wonderful to be able to have a similar code of ethics emphasis and training (perhaps we did in college, I just do not recall) for teachers. If we could implement the code down to the student level, that would be all the better.
I spent the afternoon watching online Ethics Training videos. They bring up some interesting scenarios that are not always easy to determine to immediate proper course of action. It was educational, but for the most part will not pertain to me in terms of customer contact.
The building I work in is huge and easy to get lost in. My immediate supervisor (Jen Waskow) gave me a tour of the building and even treated me to lunch in the cafeteria. She introduced me to her immediate Reports (people I am anticipating working with at least occasionally) and one of her supervisors (Michael Molacek). I will be working with Michael on one of the three statistical analyses projects they will have for Ryan and me. Note: Ryan’s first day has been delayed 1 week, so hopefully I’ll be able to get him up to speed when he gets here.
One of the projects Jen explained to me sounded very simple if all the data is corroborated easily, but I’m getting some indications that just orientating the data into a usable form may be the challenge.
Another project Jen explained to me (Michael’s project) it seems will be using Multi-vitiate data. That is just short of an AP Statistics curriculum item, so I’ll go back and read that chapter in the text book tomorrow (Chapter 14 – for some reason I remember the chapter).
Just before the end of the day, Jen suggested I Google “Business Process Metrics” to get me up to speed on some of the topics we will be looking at over the next few weeks.
Subscribe to:
Posts (Atom)