INTEGRATION OF RELIABILITY AND PRODUCT DESIGN USING DESIGN OF EXPERIMENTS (DoE)

INTEGRATION OF RELIABILITY AND PRODUCT DESIGN USING

DESIGN OF EXPERIMENTS (DoE)

Manoel de Queiroz Cordova Santos and Acires Dias

Department of Mechanical Engineering, Federal University of Santa Catarina,

Brazil ABSTRACT: Product reliability improvement depends on the adequate use of design experiments. Should this task be managed by a specific product design methodology that realizes the role of statistical designed experiments on product reliability improvement then the design process truly receives the benefits of robust design. A methodology for integrate reliability and product design methodology to improve product reliability using design of experiments is explained. This methodology involves the correct identification of reliability requirements and parameters along all the design process and how to use statistical designed experiments to improve through a response surface methodology. Once the best factor levels for the appropriate reliability parameters are determined, the use of an accelerated test to verify/ predict product reliability. The systematic use of this methodology promotes reliability growth.

INTRODUCTION

For Garving [1987], the eight dimensions of quality are: (1) performance; (2) reliability; (3) durability; (4) serviceability; (5) aesthetics; (6) features; (7) perceived quality; (8) conformance to standards. The role of reliability for an organization became a world class product quality, nowadays, is acknowledged as one of the most important of the eight dimensions of quality.

The authors formulate the following hypothesis: (1) There is a major necessity for a design methodology on Brazilian’s manufacturing industries that considers product reliability improvement using design of experiments; (2) Using this methodology really improves product reliability and helps to minimize cost; (3) The success of this methodology depends on the integration of reliability and product design processes when using design of experiment to improve product reliability.

THE ROLE OF RELIABILITY REQUIREMENTS ON PRODUCT DESIGN

Before discussing what is the role of reliability requirements on product design, one should know that Fabrycky [1990] defines the system design function as one of the producer’s tasks. The system design function is subdivided on the following categories: (1) design requirements; (2) conceptual design; (3) detail design; (4) design support; (5) engineering model/prototype development; (6) engineering test; (7) transition from design to production. Among the system

requirements the design agents should consider the expected operational life of the system and system reliability. There are several other authors as Asimov [1962], Beitz [1988], Pugh [1991], Back [1983] and Fonseca [1998] that defines how the design process should be understood. We followed the steps taken from Beitz [1998] and Fonseca [1998] for complement Fabrycky’s [1990] line of thought. On figure 1 the design process could be seen as a tool for understanding what should be the role of reliability requirements on each design phase.

Figure 1: Design process [Fonseca, 1998]

At every phase of design process there are several reliability requisites that must be considered by design agents. This effort is usually called design for reliability and the accurate translation of consumer’s needs and design parameters is of the most importance for design success. At the informational design phase the design task is decoded into needs. Fonseca [1998] says that this needs goes through a three phase study: (1) translation of consumer’s needs into consumer’s requirements; (2) translation of consumer’s requirements into design requirements; (3) translation of design requirements into design specifications. Some of this design specifications tells the design agent what reliability specifications (goals) the design should meet. Reliability goals are introduced as inputs at conceptual design phase and functional analysis, test and evaluation of design concepts (like reliability and performance factors) and robust design are some of the tasks that the design agents should perform. Reliability goals are represented by parameters like failure rate, MTBF, operational life cycle as responses of conceptual design alternatives. Through the preliminary and detail design phases the testing, evaluation and assessment of reliability becomes even more important where the responses for reliability could be tested and therefore improved. Robust design is a powerful design tool for achieving this goal and statistical designed experiments could be used for evaluation of reliability allocation.

Golomski [1995] remembers us that testing is used along all the way through the product development, but there should be two major concerns at design agent’s mind: (1) the reliability prediction was accurate, considering field experience data?; (2) There is the need for a design review or a recall based on this data? Thus an effort to improve the techniques used to produce reliability growth is needed. Considering that reliability specifications are realistic, using this information to decide which design factors affects most reliability specifications as a response could be used to improve product reliability.

This work serves the purpose to build new proceedings, technical language and techniques that could be easily understood and applied by Brazilian engineering professionals. Understanding the role of reliability on the design process is critical to achieve this purpose.

USING DESIGN OF EXPERIMENTS TO IMPROVE PRODUCT

RELIABILITY

In order to use statistical designed experimentation with engineering work, Lipson [1973] classified this experiments as: (1)experiments of evaluation; (2) experiments of comparison; (3) accelerated experiments; (4) factorial experiments; (5) sequential experiments; (6) nonparametric experiments; (7) fatigue experiments. This paper deals only with type (4) experiment, although a systemic approach should be used for optimize the design processes.

Each design experimenter has to choose from several experimentation strategies, and as Montgomery [1997] suggests, that the factorial design and fractional factorial design when used with Taguchi’s philosophy (system design; parameter design; tolerance design) and response surface methodology are the key for product specification response optimization. As was discussed, reliability specifications could be used as a response of design factors, thus it is possible to improve reliability using a statistical designed experiment. “There is value to experiments and evaluation calculations only when results differing from one’s expectations are obtained” [Taguchi, 1998]. Therefore it is expected that from an designed experiment the design agents should obtain useful conclusion about what do they have to do for implementing reliability improvements within design process phases.

Hamada [1995] had reinforced the idea that statistically designed experiments could be used to improve reliability and gave several examples from different authors and discusses a methodology to treat censored data for analyze designed experiments which requires life testing.

Condra [1993], said that, reliability is quality over time, and that the disciplines used to predict and guaranty the future and the actions taken are known as “reliability”. Thus one could categorize reliability methods by intent: (1) methods to measure and predict failures; (2) methods to accommodate failures; (3) method to prevent failures. This methods could be seen as part of a reliability spectrum, from probabilistic reliability methods to deterministic reliability methods. Figure 2 shows the expertise of reliability and the emphasis given at each method and their position on the reliability spectrum.

Figure 2: Reliability spectrum

The first reliability methods had emphasized on pure statistical analysis being on the probabilistic side of the reliability spectrum. Through the last 20 years reliability actions have been moving from methods to measure and predict failure to methods to prevent failure. The learning curve acquired from the studies of the failure modes, effects and criticality as other methods to accommodate failures set the basis for the physics-of-failure based methods. This methods are on the deterministic side of the reliability spectrum. If one takes action based on the information provided from all the reliability spectrum then product reliability is truly optimized.

Nowadays as Condra [1993] assures that design of experiments is one of the most effective of the new reliability tools for: (1) purchase reliable materials and components; (2) design reliable products, within the capabilities of the materials and manufacturing processes; (3) qualify the manufacturing and assembly processes; (4) control the manufacturing and assembly processes; (5) manage the life cycle usage of the product.

A DESIGN METHODOLOGY TO IMPROVE PRODUCT RELIABILITY

USING DESIGN OF EXPERIMENTS

Design of experiments are effective tools to reduce an experiment operational time. As Fabrycky [1990] showed, the percent of life-cycle cost committed it is imperative to “spent” as much time as possible on the earlier phases of the product life-cycle. Thus the first design phases are responsible for at least fifty percent of the life-cycle cost committed, as illustrated on figure 3. Thus, design of experiments could be used to help cost minimization through design process.

The methodology proposed for integrate reliability and product design using improvement design of experiment is based on the following set of design actions: (1) identifying realistic reliability specifications; (2) choosing appropriate factors that probably has a significant effect over the process response (that could be product reliability or an product performance that affects reliability); (3) plan the experiment using the design of experiments methodology; (4) optimization of factor levels using response surface methodology (RSM); (5) evaluate product reliability using accelerated life tests.

Figure 3: Life-cycle phases and it’s relation with cost commitments

(adapted from Fabrycky [1990])

Every design specification that could affect reliability should be listed, including systems and components that represents a future reliability major problem. The design agents should think about functional outputs, as the environmental conditions over which the product must operate, the expected operating life (MTTF) or what should be the mean-time-between-failure (MTBF) if this product is meant to be repairable. All design parameters that affects a design performance that affects reliability should be identified. As Condra [1993] said, the design of experiments are more applicable to deterministic reliability, thus design agents are recommended to develop skills on the reliability of components instead of system reliability. The studies of deterministic reliability are concerned with: (1) individual failure mechanisms and their distributions in time; (2) the investigations of causes of failure at component level; (3) understanding the failure modes at structural level thus providing means for reliability improvement; (4) distributions of single failure mechanisms which causes overstress.

On the reliability design deployment the design agents could group this specifications in two groups: (1) factors that affects reliability directly; (2) factors that affects reliability indirectly. As an example for the first group Montgomery [1997] showed some case studies as the effective life of insulating fluids at an accelerated load of 35 kV and the comparison of three brands of batteries. Both examples uses reliability specifications (type of insulating fluid; brand of battery), and both cases affects product life directly. Condra [1993] gives an example at design phase where the reliability specifications affected reliability indirectly, with the fuel rail design example it was used an designed experiment for optimize the algebraic maxima that was the response given by an FEA (finite element analysis) program. Minimization of maximum stress levels were a major problem and would indirectly affect reliability because of the reduction of the product life cycle. Thus it is indirect because the response of the experiment was stress levels, but this would have it’s own effect over reliability response. By the optimization of the factors that affect stress level the design agents made decisions that permitted reliability improvement.

Taguchi’s design philosophy uses four major concepts: (1) the loss function; (2) system design; (3) parameter design; (4) tolerance design. The loss function represents the cost paid by society when there are departure from a nominal value on production; system design is about the use of engineering knowledge for elaborate design conceptual solutions; parameter design is the specifications of

the design parameters in such a way that process variability is considered; tolerance design should be used to determine the best tolerances for the design parameters. The methodology purposed here should be used with Dr. Taguchi’s philosophy to help the reliability improvement effort.

When designing experiments one has to worry about the cost. Thus it is very important to decide how many factors are going to be used on the design and how many runs and replications (if any) one’s could afford? Which are the confidence level desired (α and β confidence levels). For example, for a 2k-p fractional factorial design Montgomery [1997] showed that at least fifteen factors could be evaluated with little effort, as we could observe on table 1. Thus a fifteen factor with two levels and two replicates for each treatment combination would require only thirty two runs.

With the results of the experiment the design agents will be able to analyze the data, set the best combination of design factors and levels. From this conclusions, response surface methodology (RSM) could be used to improve the response. If the response used was reliability related (e.g. product life) then RSM could be used to improve reliability. From the experimental data a statistical model is built and an optimization method (e.g. steepest ascent) is used to predict what can be expected to be the optimum just using the original region of experimentation, i.e., the optimum does not even need to be on the original region. Thus after the optimum life conditions are predicted a new prototype could be created and tested to assure that the predicted life was right. This could also be used to assure that the design became robust to known failure modes effects, thus improving reliability once more.

TABLE 1

K-P

2 Fractional Factorial Designs (adapted from Montgomery [1997])

Number of Design Resolution Number of factors, k Runs

3 2III3-1 4 4 2IV4-1 8 M M M

10-510-510-510-5

, 2IV, 2IV, 2IV 128, , 32, 16 9 2IV

M M M

14-10

16 14 2III

15-11

15 2III 16

After identifying the optimum treatment combination (factors and levels), the design team should conduct an accelerated life testing. Hamada [1995] developed a methodology to deal with censoring while using design of experiments to evaluate reliability as a response.

Figure 4: Purposed methodology for integrate design methodology

The methodology is purposed to integrate the design process as figure 4 describes. At every phase of the product design process the engineering effort to improve reliability using designed experiments to define the best treatment combinations, response surface methodology to optimize design treatment combination and life accelerated tests to predict product reliability. This outputs could be used as new design information inputs at every design phase and each phase could provide information for improve product reliability.

Condra [1993] gave an example of the “surface mount capacitor” and using an inner array to evaluate reliability specifications such as ambient temperature (an environmental factor) and voltage (an operating factor) both evaluated at accelerated conditions and an outer array to evaluate manufacturing parameters (dielectric composition and processing temperature). The response of this experiment was the average lifetime in hours (mean time-to-failure, MTTF). In this example no response surface methodology was made. The best treatment combination were used at two accelerated test, one for each treatment (with the respective levels). From the accelerated tests an acceleration factor was obtained and multiplying by the optimum MTTF product reliability was predicted.

CONCLUSION

The methodology presented have been enhancing the product design methodologies from a Brazilian’s point of view. It has been proving that design of experiments could be integrated at each phase of the product design process and that it helps to assure that the commitments made at product design phases are optimal from a reliability engineering point of view.

As a methodology, not all steps are required to improve reliability. As Condra [1993] demonstrated, he did not use an RSM to optimize experimental treatment

combinations. We had reinforced that Dr. Taguchi’s philosophy of robust design could be used to add more bulk to this methodology.

The success of this methodology, as considered, depends on the integration of reliability and product design processes. This integration could be made if statistical designed experiments are used to improve product reliability. A realistic approach is needed to define reliability specifications and a concurrent engineering team is preferred to achieve this goal.

There is still too much to do if an world class quality process design methodology is to be used with Brazilian’s design paradigms and as we intended to demonstrate that design of experiment has a major role on product reliability improvement. The contribution for the Brazilian’s engineering and research teams had its beginning. It’s time to improve our methods to assess and predict product reliability through all product design process. The need of integrate reliability and product design methodology with statistical designed experiments has turned into reality. The methodology provided here indicates the path to product reliability improvement from a deterministic reliability point of view and that’s an “innovation” we cannot afford to delay anymore.

REFERENCES

Garving, D. A.. Competing in the eight dimensions of quality, Harvard Business Review, Sept.-Oct, 1987.

Blanchard, Benjamin S., Fabrycky, Wolter J.. Systems engineering and analysis, 2nd ed., Prentice Hall, New Jersey, 1990.

Asimov, M.. Introduction to design: fundamentals of engineering design, Prentice Hall, New Jersey, 1962.

Pahl, G., Beitz, W.. Engineering design: a systematic approach, Spring Verlag, Berlin, 1988.

Pugh, S.. Total design integrated methods for successful product engineering, Addison Wesley Publishing Company, 1991.

Back, N.. Design of industrial products methodology, Guanabara Dois, Rio de Janeiro, 1983.

Fonseca, A,. J. H. . A systematic approach of the design requirements elaboration of industrial products design and its computational implementation, Florianópolis, Thesis proposal (Mechanical Engineering Doctor), Department of Mechanical Engineering, Federal University of Santa Catarina. Unpublished work, 1998.

Golomski, William A. Reliability & quality in design, in IEEE Transactions on Reliability, vol. 44, num. 2, 1995, 216-219.

Lipson, C., Narendra, J. S. Statistical design and analysis of engineering experiments, McGraw-Hill, New York, 1993.

Montgomery, Douglas C.. Design and analysis of experiments, 4th ed., John Wiley, New York, 1997.

Taguchi, G. System of experimental design: engineering methods to optimize quality and minimize costs, Unipub/Kraus Int’l Publ., 1998, 117-120.

Hamada, M. Using statistically designed experiments to improve reliability and to achieve robust reliability, in IEEE Transactions on Reliability, vol. 44, num. 2, 1995, 206-215.

Condra, L. W. Reliability improvement with design of experiments, Marcel Decker, New York, 1993.