Академический Документы
Профессиональный Документы
Культура Документы
Daryl L. Santos Systems Science and Industrial Engineering Department T.J. Watson School of Engineering and Applied Science Binghamton University Binghamton, NY, USA santos@binghamton.edu Srinivasa Aravamudhan and Gerald Pham-Van-Diep Speedline Technologies, Franklin, MA, USA Anand Bhosale DCI Automation, Worcester, MA, USA
ABSTRACT The purpose of this paper is to compare and contrast different stencil printing process capability results that can be obtained from the same data points, depending upon how the data are analyzed. In estimating a printing process capability to deposit solder paste, it is a common practice (see Mukadam et al. (2002) for example) to utilize process capability indices (like Cp, Cpk, and others). One set of data, depending upon how it is analyzed, can actually provide different values for the index being investigated. For example, if one is interested in the capability of the process for meeting the specifications on one particular component, different results could be obtained if the data were analyzed in the following ways: board-to-board, all deposits, parallel vs. perpendicular pad orientation (w.r.t. squeegee blade direction), and others. This paper will provide an example case study and show that one process, depending upon how it is viewed or analyzed, will in fact produce various results via process capability index analysis. Key words: Process Capability, Stencil Printing PROBLEM Many process engineers are becoming aware, particularly through 6 types of programs, of the need for a sound statistical approach to understand the capabilities of their processes and are learning many of the tools that will enable them to determine those capabilities and to subsequently improve upon them. One such tool for determining the ability of a process to meet specifications on the product is to calculate what is known as a process capability index. While the practitioner may know the formula to use in order to calculate a process capability index, like Cp or Cpk, what that same practitioner may not realize is the fact that performing a process capability calculation on a single set of data may actually yield
different results. For example, if we consider a stencil printing process and wanted to perform a process capability calculation on the ability to deposit solder for one of the components (say a QFP), then we may get different process capability results if we looked at the data from different viewpoints. In other words, if we calculated Cp or Cpk on all deposits for that component, that would likely give us a different result than if we calculated Cp or Cpk while analyzing board-to-board deposits, or analyzing by pad orientation (with respect to squeegee blade direction), etc. Based upon this concept, this paper presents a real-world case study on solder paste deposits for a QFP (one of many examples that could be used) and shows that, depending upon how that one set of data is analyzed, that very different results on process capability estimation can be obtained. Before we do that, a brief review of process capability indices will be provided. PROCESS CAPABILITY FORMULAS As companies have recently begun certifying process engineers and other employees (or sending them for training to be certified) in 6 programs (e.g., green belt or black belt certification), it should probably not be necessary to present an in-depth review of the various process capability formulas. Instead, we will present a brief review of the common process capability indices (namely, Cp and Cpk) and their calculations. A simple process capability index (like Cp, Cpk,, etc.), as many of us know, is calculated as a ratio of the tolerance of some feature (print tolerance, X-axis accuracy, placement force, etc.) to the variability of the process (as typically measured as a function of standard deviation). The two most common process capability indices that are used are Cp and Cpk. Given that USL and LSL are the upper and lower specification limits of a feature, respectively, mu ( ) represents the mean of the process, and sigma ( ) represents the standard deviation of the
65
process that is used to produce or make that feature, then the Cp and Cpk formulas are as follows:
Cp
all of size 50 (a respectable sample size), may show an average process capability (Cpk) estimate of 1.00, a good (not great) one of 1.47, or a bad one of 0.7, based on the samples being used, but the true Cpk may be none of these values! Let us now suppose, continuing the example, that the practitioner only took one sample, and it was the sample that happened to have the estimate of Cpk being 1.47. Most practitioners under that situation (and we cannot exclude ourselves from that list based upon past work) may stop there and say Cpk = 1.47 (without further analysis) when in fact that is simply not true, the estimate of Cpk may be 1.47, but we may not know, nor will we ever, the true Cpk value. However, if we place a confidence interval (CI) about that estimate of the form [LCL, UCL] where LCL is the lower confidence limit and UCL is the upper confidence limit, then we can say with a certain degree of confidence (based upon a stated alpha ( ) level) that the true value of Cpk lies within that interval. The CI is said to have 100(1- )% confidence. While alpha is usually between 1% and 10%, a typical value for alpha is 5%, resulting in a confidence interval of 95%. Because Cpk is a function of two parameters and and because there does not exist an unbiased confidence interval that has both and , then an exact confidence interval for Cpk cannot be generated (refer to Santos et al. (2004) for more discussion). This is not true for Cp. An unbiased confidence interval for does exist and therefore the unbiased confidence interval that can be used when estimates of Cp are being studied as derived exactly from the unbiased CI for is the following: Where the above is of the following form: