Improving product quality: in-depth analysis of the application of the six standard deviation DMAIC method in quality improvement (full analysis of the analysis process)
Last Modified: Jun 18, 2024 8:00 PM
If you sail against the current, you will retreat if you don't advance. Making a little better than yesterday every day is the realm of endless progress that every company hopes to achieve. This is also the highest standard followed by many high-tech semiconductor units. The quality assurance department even has a CIT (Continue Improvement Team) continuous improvement team's improvement methods, hoping to bring a culture of continuous improvement of product quality and continuous optimization of product capabilities to the company. and atmosphere.
Improving product defects and enhancing product performance capabilities are problems that everyone faces every day, which leads to many steps and methods that use standardized processing procedures and statistical analysis techniques to quickly diagnose, correct and monitor problems. The most commonly mentioned one is " DMAIC improvement method".
what isDMAICImprove methods?
DMAIC is a six sigma methodology, a powerful method designed to improve internal processes or undesirable phenomena within an organization. DMAIC represents the five key stages of the method: Define, Measure, Analyze, Improve and Control.
Using statistical analysis tools, especially JMP, whether you are a statistical master or a novice with no foundation in quality control concepts, you can easily construct stunning statistical analysis reports to help you quickly make decisions to improve your products. . Below we give a case as an illustration.
Quality problem background
Mr. Q is the quality assurance team leader of a company, responsible for ensuring that the company's product quality reaches the shipping level. Xiaopang is a fresh graduate, and by chance, he joins Mr. Q's team to work together. . Currently, the company's new product is facing a quality problem that needs to be corrected. This product needs to bond the front frame and touch screen, and the touch screen needs to be controlled to be about 0.15~0.25mm lower than the front frame. However, the current bonding effect is not good, about There is a 47% rate that exceeds the specification. When the step difference between the touch screen and the front frame is too small or the touch screen is higher than the front frame, this undesirable phenomenon will cause the user to have a poor user experience or poor touch sensitivity, or even touch failure. Control the risk of screen falling.
Xiaopang: "Currently, 47% of the touch screen and front frame are out of specification, and 27% are too small. It seems to be a serious quality problem. How can we improve this problem?"
Mr. Q: "I think this problem is relatively complicated. We may have to start a project to use DMAIC analysis techniques to solve this problem. The benefits should be great."
Xiaopang: "I have learned this concept in school, but I have never actually operated the whole process in the past. I wonder if it will be complicated?"
Mr. Q: "Don't worry, we can use JMP to complete the analysis report at each stage and get improvement measures."
Then, Mr. Q and Xiaopang entered the office to start the first phase of "definition" report preparation.
DMAIC：Definition
Mr. Q: "In the definition stage, we can use JMP's Journal, which is a unique report presentation format of JMP. It can use information such as the current production line status, problems faced or expected results to be achieved. Expressed in columnar, tabular or graphical format, it can be better combined with various JMP statistical tools.
首先，我們運用條列式的方式，將一些專案的訊息羅列於Journal中，並命名為『專案章程』，在此我們可以觀察到專案的命名、描述及目前所面臨的問題和期望獲得目標等information. "
As Mr. Q spoke, he entered the appropriate information in the Journal (Figure 1).
(Picture 1)
Xiaopang: "I have never used this function before, it looks pretty good."
Mr. Q: "The reason why I choose to use Journal as a message record is because JMP has a very good interactive analysis function . Because I am more worried about the problems caused by too small a step difference, I use the preliminary feedback data from the current production line. The graph generator of JMP draws and marks the data below the specification with different colors. You can find those points that are below the lower limit of our step difference specification (as shown in Figure 2). In addition, the bad graph is used as a point label when the cursor is used. Stopping at this point, you can also interactively display the bad phenomenon at that point, allowing the team to quickly understand the bad status of the problem and the distribution of data. "
(Figure II)
Xiaopang: "This kind of interactive data report will really make it easier for everyone to interpret."
Mr. Q: "In addition, I can also use JMP's Process screening to quickly make relevant statistical values based on needs, including values such as the out-of-regulation ratio and CPK value. You can find that not only is the out-of-regulation ratio very high, And the manufacturing process is not in a stable state (Figure 3)."
(Picture 3)
Xiaopang: "Yes, shall we move to the next step?"
Mr. Q: "You are right. In the next stage, we will collect detailed analysis data. However, before collecting data, we must first confirm that the measured data is accurate and valid."
DMAIC：Measure
Xiaopang: "How do we confirm that the measured data is valid and referenceable?"
Mr. Q: "Basically, JMP's MSA (Measurement System Analysis) platform can provide corresponding analysis. If your data is Go-No Go, Pass/Fail, etc., you can use Kappa under Attribute Gauge Chart. Analysis to confirm the stability of the measurement personnel or measurement tools. For the gap issue we are currently discussing, since it is continuous data, we can use the Gauge R&R analysis report under the Variability Gauge Chart to determine whether the measurement system is stable. ”
Mr.Q opened the Variability Gauge Chart, put the measurement personnel and measurement samples into the discussion factors, discussed the Gauge R&R analysis report of the step difference, and completed the report below.
Xiaopang: "How to interpret this report?"
Mr. Q: "We usually look at the value of %Gauge R&R , which is 7.96%. Generally speaking, if it is greater than 30%, it means that the measurement system is not good, and 10 - 30% means that the measurement system is acceptable, but the measurement needs to be improved. If the measurement system is less than 10%, it means the measurement system is very good (Figure 4).
(Picture 4)
Xiaopang: "So does this result mean that the measurement result is very good?"
Mr. Q: "Yes, and you can check the Variance Components report below. We can find that most of the variation comes from the differences between samples, which means that the measurement variation of the operator and the variation of the measurement tool contribute a lot. Small means that the measurement system can obtain effective and accurate measurement results.”
Xiaopang: "Understood, what's next?"
Mr. Q: "We need to collect more information on the results measured by these operators using these measurement tools, including information on some possible factors."
Xiaopang: "But I don't know what possible factor information to collect."
Mr. Q: "So we need help from Mr. M from the agency."
After speaking, Mr. Q picked up the phone and called Xiao M. After 10 minutes, Xiao M appeared in the office where the discussion was held.
DMAIC: Analyze
Little M started to complain as soon as he came in: "The issue of the step difference between the touch screen and the front frame is really difficult. The super rules are too high. I don't know where to start."
Mr. Q: "Don't worry, Xiaopang and I came to you just to discuss this issue together and see if we can find the key problems and good solutions."
Xiao M: "Great, because there are too many possibilities that may cause this problem. I have read the FMEA analysis report, but if we confirm all these problems, it will take a lot of time. The company should not be able to accept the time spent. So much time to deal with this problem.”
Xiaopang asked: "What is an FMEA report?"
Little M: "FMEA is the abbreviation of Failure Mode and Effects Analysis. It uses a checklist to confirm the risk of possible problems in each step, and then proposes solutions to high-risk problems to avoid producing defective products."
Mr. Q: "Little M, you mentioned a very critical issue, that is, there are too many possible causes. We must first screen out which factors are important, and then we can find the direction for subsequent corrections."
Little M: "Then how do you screen out the important factors from these numerous possible FMEA factors?"
Mr. Q: "That's why we are coming to you. We are going to perform a brain storming to build a root cause analysis diagram, commonly known as a fishbone diagram. We will first review it with our expert Xiao M to condense the possible causes. "
Little M: "You're welcome! I'll be happy if this difficult problem can be solved."
In the next half hour, the three of them listed the important factors based on their own experience and the FMEA confirmation project (see Figure 5 below).
(Picture 5)
Xiao M: "Because this bonding uses pressure glue to combine the touch screen and the front frame, the stability of the pressing fixture is quite important. I also consider many factors in this category, but after discussing it like this, it really can Eliminate many unnecessary considerations."
Mr. Q: "Yes, this is the main reason why we have to conduct brain storming discussions. Thank you both for your hard work. Next, we can start collecting information about these messages."
After several hours of information collection, Mr. Q, Xiao Pang and Xiao M gathered in the conference room to discuss.
Xiaopang: "I finally finished collecting the information. I spent a lot of time asking for information from various units."
Mr. Q: "Yes, data collection and cleaning are the steps that take the most time in the entire problem improvement process."
Little M: "With your help, compared to the headless time I spent before, this kind of time is very worthwhile, but what should we do next?"
Mr. Q asked tentatively: "Little Fatty, do you have any ideas?"
Xiaopang: "I think we can confirm the correlation of each factor to the results, see the importance of the factors, and see if we can converge on the important factors."
Mr. Q smiled and said, "Yes, this is also the main purpose of collecting this data. We can use JMP 's Response screening or Fit Y by X platform to quickly compare the correlation of each factor to the results ."
Mr. Q then used JMP to complete the report on the correlation between factors and results after several selections (Figure 6 and Figure 7).
(Picture 6)
(Picture 7)
Xiaopang: "Mr. Q, it seems that the lamination time and lamination pressure are very important. What is unexpected is that the type of pressure glue is also a very important key parameter."
Little M: "This really opened my eyes. I didn't expect that different pressure glues would cause differences in step differences."
Mr. Q: "Is the pressure time and pressure required for different pressure glues different? The compression amount of different pressure glues may be different."
Little M nodded in agreement and continued: "So now we have condensed the key issues of influence into these three key factors. Can you help me find the appropriate pressing time and pressure setting value?"
Mr. Q: "Of course, we still need help from JMP to help us set up an appropriate DOE combination and help us find an appropriate model as the basis for finding the optimal parameter combination."
DMAIC：Improve
Xiaopang: "I think it is not difficult to find from this information that different pressure glue models correspond to different pressures and lamination times. Can we find appropriate setting values according to different pressure glues?"
Xiao M: "This idea is correct. The factor settings of different pressure glues may be different. We are not sure whether our current supply of a single pressure glue manufacturer can meet our needs, so I want to observe the supply of multiple pressure glue manufacturers. The factor setting of the pressure glue will be safer.”
Mr. Q: "No problem. We can do a small experiment later to see how the settings are more appropriate. But before that, we need to confirm the possible effect terms of each factor. The designed experiment will be more effective." benefit."
Little M: "I don't quite understand what this means. Can you explain it a little more?"
Mr. Q: "Of course. Usually if we compare the correlation between factors, we can simply look at the main effect of the factor and the first-order effect of the result, that is, look at the slope of the simple regression line on the scatter plot."
He continued: "However, when the effect of a factor on the result is quadratic or high-order, or even the interaction between different factors will affect the response result, we still use the idea of looking for the main effect coefficient to design experimental collections. Data, the model derived in this way will be very different from the real situation.”
Xiaopang added: "JMP provides many DOE platforms to help us design experiments more flexibly. We do not need to meet the standard number of experiments like the previous full factor experiments, partial factor experiments, Taguchi experiments, etc. JMP 's customized experiments We can more flexibly design optimal experiments based on the number of experiments we can actually afford, and obtain experimental designs with higher CP values . We can also specify the effect types of factors according to our needs to avoid too many Experimental expenses. ”
Little M: "So if we find that some factors have interaction effects and other factors have secondary image effects, we can add the effects into consideration in the experimental design according to our needs?"
Mr. Q: "That's absolutely correct. It usually takes some time to learn different DOEs. When I was in college, I learned the concept of experimental design for one semester, and I didn't teach it completely. Besides, you have to integrate it yourself after learning. Concept, designing appropriate experiments can feel very annoying and cumbersome. However, with the help of JMP, all this can be completed in a very short time, helping you according to your needs, the number of times you can afford, and the number you want. Discuss the factor effects and even the relevant conditions between some factors to customize your own experimental design."
Little M: "Then how should we start?"
Mr. Q used the Fit Y by Regarding the laminating pressure and pressure glue factor, we can first confirm the quadratic influence under different factors. In terms of the impact of laminating pressure on the results, the accuracy of the quadratic term relative to the linear term has increased from 0.39 to 0.47. The value is about 0.08 and should be included in the model discussion. "Mr.Q then used JMP to quickly use the Fit model platform to consider the response surface effect and confirm the effect term analysis results of the regression model. (As shown in Figure
(Picture
Mr. Q: "In addition, we can also use the data collected so far to do a simple modeling and look at the current performance of different effects on the model. We can find that in addition to the lamination time, lamination pressure and pressure glue model In addition to the main effect, there is also the interaction between time and pressure, the interaction between pressure glue model and lamination time, and the time quadratic term just observed, which are all important factors that have a strong influence. The interaction profiles also show the same. Interpretation results (Figure 9).
(Picture 9)
Little M: "It seems that we have found the effect term of the key factor, then should we directly use this model to predict the factor setting value?"
Mr. Q: "My idea is to conduct an experiment to examine the possible space more comprehensively and find a more accurate model, so I suggest using a customized experiment to conduct an experiment."
As he spoke, Mr. Q opened the customized experimental platform, entered relevant information, and executed the resulting experimental combination (Figure 10).
(Picture 10)
Mr. Q: "Using JMP's DOE tool, we can quickly generate appropriate experimental combinations. Next, we need to collect data based on these combinations. Little Fatty, we should buy drinks and hire operators. We should trouble them." Help us measure the experimental data."
After a day of experiments and data collection, the three of them returned to the conference room to discuss again the next day.
Xiaopang: "I collected the data and ran a simple regression analysis based on the data. Basically, the key parameters maintain our previous judgment. (Figure 11, Figure 12)"
(Picture 11)
(Picture 12)
Little M: "That's great, and I think the accuracy of the model is also very high, with an R-square adj of about 0.98, which shows that the proportion of variation that this model can explain is very high. (Figure 13)"
(Picture 13)
Mr. Q: "Yes, and we can use Profiler to look at the prediction results of each factor combination, because we hope that the prediction results are as close to the 0.2 gap as possible, so we set the target at 0.2 and use Profiler's best Desirability value function (Profiler>Optimization and Desirability>Maximize Desirability), we can find that when the pressure glue is type1, the factor combination of the optimization result is 11.9 for bonding time and 8.4 for bonding pressure (Figure 14).
(Figure 14)
Little M: "Can I also understand the appropriate factor settings for pressure glue in type 2 and type 3?"
Mr. Q: "Yes, no problem. You can select Prediction Profiler>Reset Factor Grid to select the factor category to be fixed, and repeat the process to obtain the most appropriate factor setting values for different pressure glue models. (Figure 15)"
(Picture 15)
After some operations, Mr. Q obtained the following results (Figure 16).
(Figure 16)
Mr. Q: "I should choose type2 as the first choice, because the confidence interval of the step difference is relatively small, which means the deviation will be smaller. Different pressure glues are suitable for different lamination parameters, so the purchase volume of which pressure glue should be considered The output value can meet our expectations.”
Little M: "Understood, then I will quickly ask the manufacturer of pressure glue Type 2 if they can produce more products so that all our products can be bonded with their pressure glue."
DMAIC:control(Control)
Many days later, the three of them had the opportunity to get together again.
Little M: "It seems that the problem of step difference has been improved. Thank you very much. My ears have become much quieter these days."
Mr. Q smiled and said: You're welcome, this is the result of everyone's joint efforts. The boss should be very happy to see that the number of heavy workers has decreased, output has increased, and scraps have decreased. "
Little M: "Yes, he asked me to record the solution and then share with you how to improve and find the best combination of factors."
Xiaopang: "Then you should take a look at the control charts and ANOVA reports that have been improved in the past few days. It can clearly show that the production line's process capabilities have improved and become stable."
Little M: "Have you sorted it out here?"
Xiaopang: "Of course, please take a look at the control chart and ANOVA report I drew using JMP."
Xiaopang called up JMP's control chart report.
Xiaopang: "It seems that after fixing the appropriate pressing time, pressure and pressure glue type setting values, the variation of the step difference becomes smaller. This solution has indeed improved the bad situation of the step difference. (Figure 17)"
(Picture 17)
Mr. Q: "In addition, we can also use JMP to confirm whether there is really a difference before and after the improvement. First, we first confirm whether the distribution obeys the normal state, and use the Distribution platform to confirm the normality. The distribution before the improvement does not comply with the normal state (Fitted Normal Distribution> Goodness -Prob of shapiro-wilk of-of-Fit Test
(Figure 18)
Mr. Q then selected the Nonparametric Wilcoxon/Kruskal Wallis Test under the Fit Y by X platform and continued:
"Under such circumstances, we used a parentless method to confirm whether the average values of the two groups are equal. The result is that Prob>chiSq is <0.0001, and its value is less than 0.05, which means there is a significant difference between the average values before and after the improvement. (Figure 19 )"
(Picture 19)
Little M: "Thank you for your help. I was able to quickly solve this difficult problem and save the company a lot of money."
Mr. Q: "You're welcome, no need to thank us, thank JMP!"
Recommended reading
> Free download of "Continuous Quality Improvement: Refining Six Sigma DMAIC Analysis Guide (Chinese version)"
> Try JMP for free for 30 days to solve your problems in quality management & improvement
> Register for the Quality On-Demand Seminar to learn more about JMP’s application in Six Sigma and Quality Control
This post originally written in Chinese (Traditional) and has been translated for your convenience. When you reply, it will also be translated back to Chinese (Traditional).
Comments