Quality Strategies for Manufacturing Operations

Project Contact: Karl Kempf, Intel Corp
Proposal Contact: Tom DeNoto, Polaroid

Go to LFM Research Group 5 Mission Statement page

---

Background: Our basic understanding of quantity manufacturing is supported by our ability to be given a process flow and an equipment set along with data about machine availabilities and process cycle times, that we plug into a "model" (anything from a spreadsheet to a full discrete event simulation) to play various output volume and throughput time games. In other words, we have understood the basic interactions between the machines and processing steps and jobs sufficiently to build a generalized modeling methodology that can be applied to a variety of manufacturing scenarios.

Unfortunately, our basic understanding of quality manufacturing is hampered by our inability to be given individual Cp and/or Cpk information pertaining to the steps in the process flow and the machines that execute the processing steps, and to then utilize a common abstract modeling methodology which supports estimating overall system quality. For example, there are at least three well known approaches to quality management as popularized by Deming, Crosby, and Juran (ref. 1). However, there is no common modeling methodology that reduces these approaches to shop floor terms and allows manufacturing personnel to "simulate" and evaluate different approaches to their day to day quality problems. This proposal is aimed at beginning to provide a model (probably a specific one that might point the direction to a more general formulation) that can be used to support playing such quality "what-ifs".

Proposal Description: The research required would include four tasks:

• The first task would include detailed work with the member companies to define at least three quality "rollup" problems over multi-step manufacturing processes that are small enough to be tractable, large enough to contain all of the basic issues, and that are encountered during operation of an existing factory. In this phase, emphasis would be on understanding the events, issues, and phenomena of importance to quality in operating factories. It is expected that during this phase factors that are factory specific and factors that are common between factories would be observed. (See the Appendix for sample problems suggested by interested companies.)

• The second task would include a survey of the technical literature to find reported quality approaches which are applicable to the events, issues, and phenomena observed in task 1. It is assumed that this would include the "Deming", "Crosby", and "Juran" ideas mentioned above. It is further assumed that there would be some overlap between the methods used in operating factories and the methods described in the technical literature.

• The third task would include the construction of models of each of the problems collected in Task 1, using the information collected in Tasks 1 and 2, and the validation of these models against actual data collected in Task 1.

• The fourth task would include abstracting and combining the models constructed and validated in Task 3. The goal here is a step towards a high level formal quality management language (QuaL) which could be used to encode and evaluate different quality policies in a manufacturing facility.

Note that an obvious follow-on project is to:

• include the QuaL in the design of a manufacturing simulator,
• implement the simulator, and
• test the simulator with various policies written in QuaL against the performance of the factories surveyed to initially design QuaL.

The current proposal is to determine whether QuaL is possible. Follow-on projects might expand and implement QuaL.

Schedule: The schedule and budget would be broken into four sections, one for each task :

• Task 1: 14 weeks, visiting roughly one factory every two to four weeks, including interviews with senior manufacturing operations personnel at each site - result would be a data set for each factory including factors, metrics, and current policies as described in Task 1 above.

• Task 2: 12 weeks, including access to excellent library resources, and including brainstorming with senior manufacturing research personnel, to study the quality policies described in the literature - result would be a model of each of the policies studied.

• Task 3: 12 weeks, including brainstorming with senior manufacturing research and operations personnel, to build and validate a model of each factory, and to attempt to build and simulate, for example, a "Deming-like", "Crosby-like", and "Juran-like" policy for each factory as appropriate.

• Task 4: 12 weeks, including brainstorming with senior manufacturing research personnel, to build a common abstract representation in which it would be possible to encode all of the models built in Task 3, and to demonstrate the power of the abstract representation by reproducing the results of Task 3.

References:

• J. G. Suarez, Three Experts on Quality Management, Department of the Navy, TQL Office, Publication No. 92-02, July, 1992.

Appendix: : General Motors, Ford, Kodak, Polaroid, and Intel are willing to supply concrete examples for this Proposal as briefly described here

• At GM, tolerance stack-up methodologies (Worst-Case, RSS, Bender, Monte Carlo Simulation, ...) have been fairly well developed and are widely used by product and analysis groups to predict the "no build" conditions of the assemblies during the product development phase. The relationships required to conduct such as analysis are generally obtained from physics of the problem. However, the assumption made at this stage of design is generally that manufacturing will find ways to meet the required component feature tolerances, which in turn will result in acceptable assemblies. The problem faced by manufacturing during process selection, and in day-to-day operations, of predicting the sources of variation and their relative contribution is more complex than tolerance stack-up. Examples of the factors that should be considered include impact of serial processing on quality, mean-shifts over time, and correlation between different process parameters. It would be beneficial if the proposed research could address the above issues, identify the basic building blocks (machine, fixture, ...), and come up with a methodology to combine these elements. This methodology should be applicable to a wide range of processes, with the understanding that the impact of some of these factors may be negligible for certain situations. A tool of this nature would be of tremendous value for both plan management and process engineers, for it would help them a) predict quality, and hence enable them to conduct quality-cost trade-offs for alternate processes and system configurations, and b) troubleshoot current quality problems.

• At FORD, there are three potential areas that could benefit from a detailed quality roll-up analysis. The areas involve the production of injectors, rotor assemblies, and throttle bodies. The analysis would concentrate on a few key sections of the production lines which have been determined to be critical to the overall quality of the part, including ... - Injectors: Determine and analyze the critical parameters influencing the quality of the spray aperture. - Alternator Rotors: Determine and analyze the critical parameters influencing the quality of the rotating pole piece alignment. - Throttle Bodies: Determine the quality impact of integrating various components into the assembly process of the throttle body.

• At KODAK, photographic products result from the integration of many specific chemical components. This demands known quality at critical steps in the manufacture of all components. Most components require several steps in their manufacturing sequence. Both component attribute data (via specific tests) and manufacturing process data are available to assess quality throughout this sequence. The quality "roll up" assumes an added dimension due to chemical interactions when specific components are combined. These interactions produce the results desired in final products. This industrial example will provide information on the quality "roll up" of a cascaded chemical manufacturing operation, as well as an assessment of the impact of "time to known quality" at each step on manufacturing strategy. This example will be limited to the manufacture of selected key components and their interactions, not final product assembly.

• In manufacturing at POLAROID, there exists an issue in determining the appropriate measurement/display parameters and feedback loops and their linkage to process measures and control (SPC) and to final products variation and, therefore, quality. The methodology is uncertain and, therefore, the appropriate and optimum number of measurements are unknown which results in the non-optimal deployment of resources both in measurement and test design. Specifically, Polaroid has a number of complex processing machines that potentially have hundreds of control points and a number of intermediate steps in which partial product tests may be performed. A methodology for selection and linkage of these information points is essential for excellence in the manufacturing of complex, technically intricate products. There is a pilot opportunity in both hardware manufacture and film component manufacture to validate the methodology.

• At INTEL, tolerance stackup in wafer fabrication is a complex problem. The process is hundreds of steps long, with many steps interacting over long distances in non-linear ways. The basis for the interaction lies in the process physics, some of which is well-known, some of which is still a black art. Numerous local examples are available for study, results from which would be of benefit to Intel and the student. Quality management is also a complex problem in die assembly and test. Part of the problem is in assembling die from inside Intel with packages from outside vendors. Part of the problem is in the rollup of the quality of each process step in the assembly/test line into the quality of the final product as delivered to the customer. Again, numerous examples are available for study in operational factories.