Performance Verification Throughout the Product Life Cycle Using Accelerated Life Testing

Transcription

1 Performance Verification Throughout the Product Life Cycle Using Accelerated Life Testing Sarath Jayatilleka, CRE, Daikin Applied Americas Key Words: Product Life, Design Life, Accelerated Life Testing SUMMARY & CONCLUSIONS The term Life in an Accelerated Life Testing (ALT) is a key component. For an unknown product design life, an ALT cannot be designed. The reliability requirements are about the probability of satisfying specific product performance parameters across the product life cycle. Therefore, defining the design life of a product becomes a major requirement in designing an ALT. Expected product life at service in the market place is mostly defined by the customers in their viewpoint and time scales such as years. Customer s perception on product life is influenced by the market forces. The techniques applicable to the process of translating voice of the customer into a well-defined product life and product characteristics are discussed briefly with examples. Once the expected product life in service is defined in customer s terms, it needs to be translated and defined in engineering terms at the product (system) level. This would be called the product design life. A 10-year product design life of a washing machine can be translated into number of washcycles across 10 years. Then the number of washing cycles in 10 years becomes the system level design life in designer s terms. More elaborate examples from appliance, medical device and heating, ventilating & air conditioning (HVAC) field are discussed. Washing cycles of a household clothes washer has fewer seasonal functional usage duty cycles (FUDC) while HVAC equipment s have highly seasonal FUDCs. In today s and future trends, cloud based equipment can provide well-defined FUDCs. Some FUDCs are easily derived while others need complex models and research. Methods and examples are discussed for both simple and complex scenarios. Systems are made of subsystem and component integrations. Suppliers are involved in supplying different subsystems and components. Supplier reliability requirements that communicate design life to suppliers are not generated in customer s language such as years but in component FUDCs. Therefore, system level design life requirements are decomposed to subsystem and component level FUDCs. A wash cycle of a washing machine can be decomposed into a bearing design life in terms of number of bearing revolution with specific radial and axial loads in a defined humid environment. The process of reliability requirement and product design life decomposition is discussed using a Systems Engineering V-Model. The product design life requirement derivation is a simple process in some cases. A design life of the gear wheel attached to the shorthand of a clock is a simple example. It becomes challenging in an insulated-gate bipolar transistor (IGBT) of a variable frequency drive. Both simple and complex processes are discussed with examples. Finally, the benefits of welldesigned ALTs for the purpose of performance verification across the product lifecycle are discussed with examples. 1.1 Design Life 1. RELIABILITY AND DESIGN LIFE The design life of a product (or its components) is the period of time during which that item is expected by its designers to perform intended functions within its specified design parameters and operational environments. Once the design life is known, it can be incorporated as a key design input to the design system. The process of incorporating design life into a design system is an important activity within design for reliability. Market research is required for determining the customer s expected design life. Unlike a pure science such as reliability engineering, market research is a social science. A typical indirect question a market researcher for washing machine industry would ask a consumer is, If your friend has a choice over the product life, what years of product life would you recommend for your friend who is looking for a washing machine for his newly built home? In the place of more technical term design life, the less technical and more social term product life is used in market research. Typical design life of a design system is defined in 5 year multiples. If competitors are making a similar product for 20- year design life, competitive strategy is to make a product for 25 years or compete in the market place with 20 year life product with better reliability. Reliability is a competitive edge for brand marketing. For example, an automobile company may advertise 90% of a certain brand product is still out on the road after 10 years. Today s consumers have a perception about the expected product life. For example, most home appliances are typically intended for a 10-year design life. With such information, a company s marketing group may determine the expected design life of the new product. 1.2 Product Life in Service Product life in service is the length of time between /17/$ IEEE

2 placement into service of a product and that item s onset of wear out. In other words, it is the time to show up the first design deficiency in the field. Some design issues appear after the five year warranty period. Customer oriented organizations implement design for reliability into their new product development in order to extend the product life in service beyond design life. A misconception about the preventive maintenance (PM) is that it can extend the product design life. Instead, PM is an integral part within the design to meet the design life. PM is required to achieve design life in service. In order to highlight design issues, field reliability data of a certain electronic component in an industrial system is illustrated using a Weibull Plot in Figure 1.1. The full population data set including time to failure and time to survival was available. In the Weibull plot, failures were not discriminated by failure modes. There are three categories of data that are circled. The DoA category indicates the parts that were dead on arrival. The data before the 365 day line are due to manufacturing related issues. These data fall into infant mortality. Proper failure analysis was conducted to verify they were manufacturing issues. The third data set is due to design issues. The time to show up this issue is around 1000 days. Therefore product life in service is approximately 1000 days. Figure 1.1: Field Issues of a an Electronic Component 2 DERIVING THE DESIGN LIFE After the expected design life is derived from the voice of the customer, next step is to derive the system design life in engineering terms or FUDCs. As an example, a 10-year design life of a washing machine can be defined in number of washing cycles. In this case FUDC is the washing cycle. An example of how to derive the number of washing cycles that a washing machine would run in 10 years is explained herein. In more networked, cloud based industries such data could be readily available in real time. A household survey in order to obtain user data can be designed as follows. Select 100 household participants stratified around the country who agreed to participate in this program. Provide them with weighing scales, clothes basket and a document to record every clothing load weight before loading. Record the number of households at each home and their ages. After 30 days, collect data sheets. Clothe loads from 100 household can be used to develop load and wash cycle usage profiles. The number of washing cycles per 30 days can be ranked large to small and the 90 th percentile value can be normalized to washing cycles per 10 years as the design life. This typically means 90% of the consumer s use rates are considered in defining the design life. While 10% of the consumers may experience less than 10-year design life, 90% of the consumer population will enjoy the design life beyond 10 years as shown in Figure 2.1. Figure 2.1 Usage Probability Density Function Another simple example would be a blood analyzing system s design life. A 10-year design life of a blood analyzer can be defined in number of blood tests it performs. For the blood analyzer FUDC is the blood test. There are over hundred types of blood tests that can be performed by an analyzer. Types of tests are prioritized based on the patient s urgency. Tests are chemical agent dependent and less mechanical part dependent. Medical device industry is regulated by Federal Drug Administrations. Therefore, blood analyzers may be connected for real-time performance monitoring. Such cases are more common and provide details into failure modes, service conditions and load levels. If the number of blood tests performed during the past 365 days can be found for networked blood analyzers across North America for 100 samples, the 95 th percentile can be used as the North American design life in terms of number of tests. The top 5% of use-rate customers bring more sales through chemical agents and PM services. They can be considered in a special customer category and offer special preventive maintenance and design for reliability packages. Special packages may include more durable materials, special bearings and price packages in addition to a new design life. A complex example would be the design life of an insulated-gate bipolar transistor (IGBT) in a variable frequency drive (VFD) that drives a compressor motor. A compressor drives the thermodynamic cycle of chillers in HVAC industry. The US Department of Energy (DoE) provides several guidelines and helps with energy models. A model called EnergyPlus was used to generate Compressor Power vs Time in 5-minute intervals for a chiller of 500 Ton capacity performing in a building in Houston, TX across 365 days. Houston, TX was selected to stay within the 90 th or 95 th percentile user concept considering following reasons. It is in the DoE weather region 6 (region 7 being the extreme warm)

3 and also the 4 th largest (by population) city in the USA. From the Power vs Time profile resulted from EnergyPlus model, compressor motor shaft Speed vs Time curve can be generated using affinity laws; Power α (Speed.) 3 Motor name plate data are also required for this analysis. An example Speed vs Time curve for compressor in 500 Ton chiller created for a particular hospital building in Houston, TX weather is illustrated in Figure 2.2. When the annual Speed vs Time profile is available, annual switching frequency change profile within an IGBT can be derived. When the switching frequency profile is available, thermal cycle profile within IGBT solder joints can be established. During thermal cycling IGBT solder joint failures often are caused by the thermal fatigue due to mismatch in coefficients of expansions of solder joint materials. IGBT manufacturers have established Switching Frequency vs. Temperature Increase (ΔT) curves for solder joints. Thereby IGBT design life in number of thermal cycles can be established. systems engineering principles. Further, systems engineering integrates all the disciplines and specialty groups into a team effort forming a structured development process that proceeds from concept to production and to operation. Systems engineering considers both the business and the technical needs of all customers with the goal of providing a quality product that meets the user needs. The Systems Engineering V-Model (Figure 2.3) is briefly described herein. It helps visualizing decomposition of requirements and verification & validation (V&V) of meeting requirements as the integration progresses. More descriptively, it is a system development model that; Reflects concept of requirement decomposition and integration Displays relationship of verification planning to requirements development Emphasizes in-process validation Emphasizes continuing opportunity for risk management through early problem identification and resolution Illustrates impact of time and maturity Figure 2.2 Speed vs Time Profile of a Compressor Motor There are other design lives that can be derived from already available information for a compressor of a chiller. For example, components of oil pump design life attached to the compressor in the above example can also be derived. All the bearing in the oil pump are of hydrodynamic type. The most damaging wear cycle happens due to metal to metal contact within the bearings when the motor is starting up. Compressor main shaft is directly coupled to the gear pump. Gear pump is located at the sump of the compressor. Using the compressor shaft speed profile for a gear pump, the number of stops can be obtained. This is the bearing life in start-stop FUDCs. Since the oil pump is a geared pump then gear wear life depends on number of revolutions. The area under the Speed (rpm) vs Time (minutes) curve provides the number of revolutions per year. Area under the curve multiplied by 10 years provides the gear expected design life in revolutions. 2.1 Systems Engineering V-Model Meeting essential requirements across the design life is paramount to product reliability. This concept is embedded in Figure 2.3 Systems Engineering V-Model In order to explain the V-model, a home refrigerator is considered herein. When the customer needs (located at top left of the V-model) his/her food preservation for a reasonable length of time, a refrigerator system provides the solution by maintaining food at a certain temperature range to support food preservation. This is the system requirement in the V- model. Down in the V-model, a module would be the compressor unit that performs to meet system level requirements. The reliability requirement for this module could be as follows. The compressor shall have 99.9% reliability at 95% confidence when performing its intended function in the household environment across 10-year design life. This requirement is further decomposed down to several other requirements. As derived under the item (b) of section 2.2.1, the compressor reliability shall be, R (175,200 start-stop cycles) > 99.9% at 95% confidence. A conservative example verification method would lead to ALT conditions such as running 15 compressors 250,000 start-stop cycles (approximately 140% of the design life) with zero failures.

4 2.2 Deriving Subsystem & Component Design Life When a design engineer sits down to design a life into a product or its subsystems and components, design life in customer s language such as 15 or 20 years doesn t help him much for his engineering background. If the design life of a spring is given in number of compressions with its elongation limits that the spring would excite during a 20 year design life, an engineer now can perform the engineering design analysis within the safer zone of the fatigue endurance limit. This specification also can be transferred to suppliers and design test plans to verify quantified spring reliability at the end of design life. This process of translating the top (system) level requirements, (e.g. 20 years) down to component level functional usage duty cycles (e.g. Number of spring compressions in 20 years) represents decomposition of requirements in systems engineering discipline. Table 2.1 Examples of System and Component FUDCs System Subsystem/Component System FUDC Component FUDC Table Clock Hours Gear wheel Revolutions Washing Machine Wash Cycles Bearing Revolutions Automobile Miles Tire Miles Ball Point Pen Feet of writing Spring Compressions iphone Years Home Button Presses Electric Motor Hours Hydrodynamic Bearings Start-stops Here the start-stop cycle was considered as the most damaging cycle to a compressor and to the motor that drives the compressor. Refrigerator Door: Similar research found that the refrigerator door would be opened and closed by a 90th percentile extreme household user 24 times a day. Then the lifespan of the door hinge is; 24 open-close cycles/day x 365 days/ yr x 10 yrs = Lifespan of open/close cycles. Here we considered the open-close cycle as the most damaging cycle to a refrigerator door hinge. The functional usage duty cycles such as start-stop of a motor or openclose cycle of a door hinge shall be the FUDC unit of design life at subsystem/component level in engineering specifications. These specifications can be transferred to suppliers through engineering drawings as contractual agreements. In return, companies expect objective evidence such as Weibull plots from suppliers for meeting reliability requirements; R (87600 open-close cycles) > 99.9% at 95% confidence. 3 DESIGN LIFE VERIFICATION USING ALT A perfect design in the first design iteration may not practically happen. In new product development (NPD) process, subsystems and components often discover failure modes in their early life stages. As the NPD progresses across the design phase and systems are improved early failure modes gradually fade away. This can be illustrated in reliability plots and described in ref [1.] When the design life is known, how far a design fell short of the target can be illustrated in a Weibull Plots similar to Figure 3.1. Thermostat Hours Relay Actuations Design Life Requirement Decomposition Examples: (a) Table Clock: In this example, a table clock is the considered system. If the design life of the table clock is 10 years, then the life of the gear wheel (component) attached to the long hand (component) is; = 1 rev/min x 60 min/hr x 24 hrs/ day x days/year x 10 years = 5,259,600 revs Here, one revolution is considered as the FUDC for the gear wheel as the most damaging cycle such as gear wear that can create backlash and display errors in time. (b) Household Refrigerator: In this example a household refrigerator and its compressor were considered as the system and its component, respectively. A household refrigerator s design life spreads across 10 years working 24 hours every day. Compressor: Research showed that the compressor of the refrigerator ran for only a few minutes, say total 2 minutes for every 30 minutes. That means, compressor design life in start/stop; 2 start-stops/hr x 24 hrs/day x 365 days/ year x 10 years = Lifespan of 175,200 start-stops. Figure 3.1 A transmission life progress across NPD Those failure modes that forced the product or components to fail early can be designed out by design improvements. By the time the product reaches maturity, there would not be failure points remaining to appear on a reliability plot. Another feature that can be observed in the linearized reliability plots is that the lines representing the design iterations tend to become steeper and get parallel towards

5 maturity. When a subsystem is incrementally improved, the causes of failure may be delayed and hence the failures may present much later. This means that the same failure may present after the expected design life. Even though failures are delayed, their patterns and distribution shapes tend to remain the same. The same distributions typically will have the same shape parameter. The slope in Weibull reliability plot represents the shape parameter. Therefore, it is reasonable to use the shape parameters from the previous parallel or almost parallel plots to verify design life using ALTs when the designs are improved to the point of zero failures. These types of test-fix-test iterations and design improvements can be successfully implemented until design life is exceeded at component and module levels. Module level and subsystem level tests are encouraged against component level tests. Module level tests better represent the customer needs. Also, with the sheer number of components in a system it is sometimes prohibitively expensive to run many individual tests. Therefore, a bearing assembly subsystem testing is encouraged against individual bearing testing. Such a bearing assembly test not only helps determining the bearing assembly reliability, but also the individual bearing reliability. 3.1 Why ALT? The goal of ALT is the reliability estimation at the user level. It also uncovers the relevant failure modes in their natural sequence. The design life of each design iteration is estimated using ALT during the NPD. ALT provides knowledge how far a particular design configuration fell short of the required design life and helps prioritizing next level of failure mode elimination or mitigation in order to meet the reliability requirements. ALTs can be categorized into two major types. The first type has an acceleration factor (AF) of unity. Many products are not in continuous use. For example, a household refrigerator works all day long for 10 years, yet its compressor works a fraction of the time, say 2 minutes per every 30 minutes, as considered in the example in item (b) of section The strategy in designing this type of ATLs is to eliminate the non-value-added portion of the duty cycle. While keeping the user level stresses, focus is on elimination of nonstressful, non-damaging cycles and waiting times of the duty cycle to concentrate only on the damaging cycle. This ALT type is also known as the time compression type. 3.2 Design Life Verification using ALT (AF=1) The Figure 3.2 shows the Weibull Unreliability Plot for an input shaft design of a certain home appliance. The input shaft FUDC is considered as shaft revolutions under certain radial and axial loads at the bearings and the shaft end pulley. To accelerate the FUDC accumulation, shafts were run faster than the user level with same radial loads. The X-axis in Figure 3.2 is proportionated to hours of operation. There had been three design iterations as shown in three lines on the plot. In the first design configuration, the shaft was supported with one sleeve bearing and a needle bearing. The shaft had a sharp fillet at the bottom and was not hardened. At the final stage, the bush was changed to a ball bearing and the needle-bearing journal was hardened. Also, the fillet diameters were improved. This ALT was conducted with simulated loads as in operational levels, with ten times faster than the normal operational levels. This is an example of a time compression type of ALT. The three failure modes are namely bearing failures, shaft wear and shaft fatigues. Most importantly, last design iteration failure point has surpassed the design life target. Later more production intended parts can be tested longer to improve on confidence limits. Figure 3.2 Test Results 3.3 Design Life Verification using 3-Stress ALT (AF>1) This example has the acceleration factor above unity. The rubber power transmission belt considered here has a failure mode of fraying due to reverse bending in its application in clothes dryers. A test fixture was designed to create reverse bending of the belt. Belt tension is controlled through a hanging pulley load. In this test, three parameters could be used for acceleration. Belt speed as in this case was used at user level. Belt fraying due to reverse bending was the targeted failure mode. Therefore, there were four points of similar reverse bending in the test fixture. Pass-fail criterion is to identify signs of belt fraying during the periodic belt inspections. User level belt tension was at 40 lbs. Three higher belt tensions were used in the ALT. Typical higher levels could be 60, 80 and 100 lbs. Design life at 40 lbs is translated into operation hours (and multiplied by certain factor for publication purpose.) The test data were used to build the Log (life) vs Log (stress) curves in Figure 3.5. The design life below the lower 99 percentile line verifies exceeding 99% reliability. In addition, the method used in reference [2] can be used to determine the reliability at the end of design life. The 3-stress level accelerated life test is described in reference [3.] Combining several parameter accelerations in one ALT has to be carefully performed. In general, true acceleration factor is more than the multiplication of the actual parameters. There are published models to estimate acceleration factors. Acceleration factors obtained in real tests can be used for better estimations and design life verifications than model

6 based acceleration factor predictions. In reliability labs, test fixtures have been used over time, bugs fixed and wellcorrelated to the actual applications. Therefore, wellestablished real-life to fixture-life correlation factors are part of the test lab tribal knowledge. Figure 3.5 Belt test results at three tresses 4 FUTURE WORK IN DESIGN LIFE INCORPORATION Design Life incorporation into designs is presently a weak area in Design for Reliability applications. The nearest thing to reliability taught in a typical college mechanical engineering degree would be the safety factor and fatigue limits. Safety factor is a necessary condition for reliability, yet an insufficient factor. In order to close this gap, incorporation of the design life into design can be accomplished within the nation s education system through already existing college courses such as Machine Designs, Strength of Materials and Mechanics of Machines. Such design for reliability incorporation into college courses is a great way to equip next generation engineers who would build next generation instruments, medical devices, automated systems and infrastructures of high reliability. The most expensive way to find unreliability is to find it in the field. Design for reliability is a cost-effective value proposition. Reliable products made earth friendly is the best thing for the environment. Therefore, incorporation of design for reliability with design life emphasis into college courses is of paramount importance today. REFERENCES 1. S. Jayatilleka, J. Okogbaa, "Use of Accelerated Life Tests for Speedier Product Development: Problems and Strategies, 51 st International Reliability Symposium on Product Quality & Integrity, San Diego, CA, pp S. Jayatilleka, J. Okogbaa., "Use of Accelerated Life Tests on Transmission Belts for Predicting Product Life, Identifying Better Designs, Materials and Suppliers," 49 th International Reliability Symposium on Product Quality & Integrity, Tampa, FL, 2003, pp W. Q. Meeker, L. A. Escobar., Statistical Methods for Reliability Data, John Wiley Series, 1998, pp BIOGRAPHY Sarath Jayatilleka, ASQ CRE, MSE, MBA. Corporate Reliability Engineer DaikinApplied Americas, 13600, Industrial Park Blvd, Plymouth, MN 55441, USA Sarath.Jayatilleka@gmail.com Sarath Jayatilleka is the Corporate Reliability Engineer at DaikinApplied Americas. He was the winner of Prestigious IIE - William Golmoski Award from 52 nd Annual Reliability and Maintainability Symposium. He has published over 8 papers in leading conferences. He had helped improving Weibull Analysis in leading commercial Statistical Program. He holds two masters degrees in engineering and business administrations. He is a senior member of American Society for Quality (ASQ,) member of IISE, IEEE and an ASQ Certified Reliability Engineer. He was the ( ) past president of the SRE Twin Cities Chapter, Minnesota.