Whether it is OEM professional manufacturing, ODM original factory commissioned design, or even outsourcing, the modern industry has one thing in common. So what is this common point? First of all, the factory produces products, and the most important thing is the yield of the product, which is the percentage that falls within the specifications. However, if you want to evaluate the quality of the product, the yield is not the only indicator. The most common indicator is the quality of the process. Such as CPK>1.33, etc. However, we are familiar with large companies such as Google, Intel, Apple, etc. Why can they make product quality stable? This involves the tools and indicators they use for evaluation. The following will introduce a simple SMT case and use JMP to conduct it. analysis.
In the SMT (Surface–mount technology) process production line, one of the stations' glue process requires evaluating the quality of equipment from two different manufacturers to select the appropriate machine. Before the analysis, let us briefly understand the dispensing process.
In the dispensing process, the thermal expansion coefficients of the wafer and the substrate are different. Temperature changes cause thermal stress to concentrate on the bumps between them, causing cracks and open circuits (as shown in Figure 1). When the substrate is deformed by external force or the environment is oxidized, the bumps will also become deformed. It is brittle and deteriorates, so it is necessary to apply glue to increase the joint strength between the chip and the substrate, isolate the air and redistribute the stress (Figure 2). Of course, there are many factors in the glue dispensing process. This time we only simply Evaluate the weight consistency and deflection in the x and y directions after dispensing on the machine.
Picture 1 on the left; Picture 2 on the right
The technical specifications usually provided by the machine are shown in Figure 3 below (not the standard for this example)
image 3
01: Narrative statistical analysis comparison
When analyzing, we first start with the dispensing weight. The data collection includes company (M, W) and number of days data (Figure 4 below). It is recommended to start with narrative statistics To get started, first understand its distribution and enter the specifications for comparison.
Figure 4
You can see from (Figure 5 below) that neither company has a very symmetrical distribution, which can be easily compared using the Distribution and Graph Builder platforms in JMP.
Figure 5
Then combined with the normality test (as shown in Figure 6 below), it is determined to be non-normal allocation , so we can then compare based on process capabilities,
Figure 6
The process capability of Company M is higher than that of Company W, but both are higher than 1.33 (Figure 7)
Figure 7
02: Hypothesis Test
After comparing the narrative statistics, you can then use the hypothesis test to understand whether there is a significant difference between the two companies. From the test results, you can know that the variation between the two is not equal (because both are non-normal distributions, use Levene's method to judge). Then from the Wetch Test, it is known that there is a significant difference between the two means, so in the weight part, it can be clearly seen that Company M has an advantage, as shown in Figure 8.
Figure 8
03: ANOVA analysis
The next step is to focus on the number of days to understand whether different days will cause significant differences in the weight of the two companies. Based on the collected data, conduct ANOVA analysis on different days, as shown in Figures 9 and 10 below. Among them, Company M has a variation in the two days There is no significant difference in some parts, but there is a significant difference in the mean part. Assuming that the deviation does not affect the quality, the general company will set another Practical difference as the level to see whether it will actually affect the quality. It is obvious that the levels of W Company are significantly different between the two days. Even if the mean is not significantly different, generally speaking, adjusting for variation will be more difficult than adjusting for deviation.
Figure 9 – ANOVA analysis of Company M
Figure 10 – ANOVA analysis of Company W
04: Control Chart
In addition to statistical testing, you can also use the control chart function (Figure 11) to filter and compare data with the Local Data Filter. It can be seen from the graph that the control boundaries of the two equipment manufacturers on the second day are narrow, but it is obvious for W Company There are points that are close to the specifications. If it is really due to price considerations, you must investigate the reasons.
Figure 11
Then confirm the degree of deviation of There are significant differences among equipment manufacturers.
Figure 12
Figure 13
Then it is divided into two days for comparison (Figure 14). Only the differences are listed here. W Company will have differences in the number of days regardless of the x or y direction, while M Company will have no impact.
Figure 14
Finally, check with the control chart (Figure 15) to confirm the level of W company. M company will not put the above picture unless there is a significant difference. Therefore, whether it is the point intersection weight or xy offset confirmation, M company is better.
Figure 15
in conclusion
Finally, from the client's perspective, there are many quality parameters to confirm for just one manufacturing process. In addition, each parameter also has different methods to analyze. Although the above functions are scattered on different platforms, they can be customized with JSL The window or Dashboard gathers and standardizes all the analysis results, echoing the common point mentioned at the beginning, which is "quality and timeliness of analysis". There are many applications in different industries in real life. Welcome to sign up for JMP online seminars Yes, you will hear more practical analysis scenarios to help you solve analysis problems in daily work.
Latest online seminar // 6 practical applications of data science in high technology
In complex high-tech production lines, to effectively improve product competitiveness and optimize processes, using data to discover, solve and prevent problems has become an important key that cannot be ignored. This seminar will start from the perspective of production, through the sharing of practical application cases, understand how JMP can effectively reduce unnecessary working hours and improve work efficiency; then learn and avoid losses, try to use the most practical examples and the most efficient methods Methods to solve and discover problems, further optimize production conditions and enable early warning and timely decision-making. This is a free seminar. Click here to register now
