Load characterization during transportation - Semantic Scholar

Microelectronics Reliability 44 (2004) 333–338 www.elsevier.com/locate/microrel

Load characterization during transportation Arun Ramakrishnan, Michael Pecht

*

CALCE Electronic Products and Systems Center, University of Maryland, College Park, MD 20742, USA Received 16 April 2003; received in revised form 23 May 2003

Abstract The reliability of a product is a function of the product’s architecture, constituent parts and materials, as well as its life cycle environmental conditions. Of these, determination of the life cycle environment (which includes manufacturing, handling, transportation, and application conditions) is often the most uncertain part in the calculation of reliability. This paper presents a methodology to measure and characterize the dynamic nature of a transportation environment induced by a commercial carrier service on a packaged product. An example is given, and the measured environment is compared with standard industry data. 2003 Elsevier Ltd. All rights reserved.

1. Introduction Reliability prediction and assessment specifications for electronic parts and products can be traced back to November 1956, with the publication of the Rome Air Development Center’s RCA Release TR-1100, ‘‘Reliability Stress Analysis for Electronic Equipment.’’ The TR-1100 document was followed by the release of MILHDBK-217, ‘‘Reliability Prediction of Electronic Equipment’’ and progeny, including Bell Communications Research’s TR-332, the Society of Automotive Engineers’ PREL, the Reliability Analysis Center’s PRISM, Siemens’ SN29500 standard, and British Telecommunications’ HRD-5. Today, most of these handbook methods have been rejected and replaced by more science-based methods, including similarity analysis, stress and damage analysis (such as physics-of-failure), and field and test-based prediction analysis [1–4]. Reliability prediction typically involves four steps: defining and modeling the product architecture and material properties, defining and characterizing the life cycle environment, computing the stress profile on the product,

* Corresponding author. Tel.: +1-301-405-5323; fax: +1-301314-9269. E-mail address: [email protected] (M. Pecht). URL: http://www.calce.umd.edu.

and using the computed stress to assess failure mechanism damage and determine the probability of failure [5]. Among these steps, defining the life cycle environment has received comparatively less attention, and can be an uncertain part of the reliability prediction process. Standard industry reliability prediction procedures take the form of an environmental test or an operational profile that must be withstood prior to a product’s being released for use. However, most industry tests have little relation to the actual life cycle environment, and instead focus on extreme operating conditions, design or destruct limits of operation, or standards developed according to industry consensus. As a result, many standard industry tests do not evaluate the actual reliability of a product, but only demonstrate to customers and regulatory agencies that the product will operate under some test conditions, which may or may not satisfy the actual conditions seen by the product. IEC 60134 [6], ‘‘Rating systems for electronic tubes and valves and analogous semiconductor devices’’ states that, ‘‘[A part] should be designed so not to exceed the absolute-maximum value for intended service under the worst probable operating conditions with respect to supply voltage variation, equipment component variation, equipment control adjustment, load variations, signal variation, environmental conditions, and variation in characteristics of the device.’’ The part manufacturer selects the absolute maximum value, and the equipment designer is responsible for ensuring that this

0026-2714/$ - see front matter 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0026-2714(03)00189-6

334

A. Ramakrishnan, M. Pecht / Microelectronics Reliability 44 (2004) 333–338

value is not exceeded during application [7]. The only way of ensuring this is to actually measure the application life cycle environment. The actual environmental loading on a product for reliability assessment is often assumed based on engineering specifications or conjecture, rather than measured. If the actual life cycle loads are different from the assumed ones, this approach can lead to costly overdesign or hazardous underdesign, and consequently, increased investment. However, if the actual life cycle environment for an product can be measured, new products for the same environment can be designed for the loads they are expected to encounter, leading to more practical designs. One of the constituents of the product life cycle environment is transportation. A typical transportation environment is a multi-load environment, including vibration, temperature, and humidity cycles, which may subject certain parts (e.g., parts intended to operate in office or home environments) to loads higher than their designed values. Accelerated tests must be tailored to cause sufficient stress to simulate the life cycle environment, but not enough to cause unrealistic failures [8]. Thus the main objectives of this paper are to demonstrate a practical methodology to measure the environment experienced by a product during transportation, and to define an approach to characterize the measured data.

2. Vibration and shock measurement Vibrations result from dynamic forces, which set up a series of motions within a system [9]. The motion may be linear, angular (torsional), or a combination of both linear and angular motion. Vibration can be measured using several instruments, such as accelerometers (e.g., electrostatic and piezoelectric), velocity sensors (e.g., magnetic transducers and tachogenerators), and displacement sensors (e.g., proximity switches and strain gages). In general, accelerometers are preferred over velocity and displacement sensors for vibration measurement for the following reasons [10]: • Accelerometers have a wider frequency range than velocity and displacement sensors; • Dynamic events (such as transients and shocks) can be more easily measured by accelerometers than velocity and displacement sensors; and • Integration is preferred over differentiation in electronic circuitry (a time derivative is usually corrupted by high-frequency noise). Hence, displacement and velocity can be obtained by simple integration of acceleration by electronic circuitry. Most of today’s commercial accelerometers are based on the piezoelectric effect. In piezoelectric sensors, a strain associated with the deformation of a sensing mass

is converted into electrical charge by a piezoelectric crystal (usually lead-zirconate titanate or quartz). A piezoelectric crystal produces an electrical output only when there is a change in the applied force. In mathematical terms, the electrical output of a piezoelectric crystal as a function of the applied force can be expressed as [11]: V0 ¼ K

dF dt

ð1Þ

where V0 is the output voltage (V), K is a proportionality constant, dF is the change in the applied force (N), and dt is the time interval over which the change in force took place (s). Eq. (1) shows that if the change in the applied force is zero, the voltage output (V0 ) is also zero. Further, if dF is constant over a long period of time (dt ! 1), V0 is still zero. However, in the case of vibration or shock, the acceleration of the sensing mass changes rapidly, making piezoelectric sensors an excellent choice for measuring these variables. During vibration, the sensing mass experiences rapid and cyclic changes in its velocity direction, i.e., the mass first moves in one direction and then in the other. This motion implies that for an instant, the velocity of the mass is zero, and the mass then acquires a velocity value in a direction opposite to its original direction of motion. This change in the velocity of the mass over a finite time produces an acceleration that is sensed by the piezoelectric crystal. During shock, the sensing mass experiences a sudden change in its velocity, which produces a large deceleration in the direction opposite to the original velocity direction, which is sensed by the piezoelectric crystal [11].

3. Background of the experiment Vibration environments have been typically defined through industry standards, published literature, or, if available, through field measurements. For example, according to MIL-STD-810F, a commercial U.S. highway truck vibration environment can be represented as shown in Fig. 1. Fig. 1 shows the power spectral density curves (to be defined later) for vibration in the vertical, transverse, and horizontal directions. However, no information about the magnitude or duration of the vibration is provided. In order to find the actual loads experienced by a typical package (in this paper, a computer monitor) during transportation, the authors carried out an experiment in which a computer monitor was shipped from College Park, Maryland to San Pedro, California and back via Federal Express. A self-powered and selfcontained environmental data was used to record the


335

Fig. 1. US highway truck vibration [12].

exceeded before that parameter is recorded. This feature is primarily for measuring dynamic data. • Time-triggered: This option allows the user to specify the minimum time interval to elapse before a parameter is recorded. This feature is primarily for measuring static data.

Fig. 2. The shock and vibration environment recorder (SAVER).

The user selects the frequency range and the number of analysis lines for dynamic data analysis, which is used to fix the sampling rate (i.e., the number of samples counted per second) and the sample size (i.e., the number of samples per record). Given the maximum frequency of interest and the number of analysis lines in the power spectral density plot, the recorder software calculates the sampling rate and sample size as follows [13]: Maximum frequency ðHzÞ ¼

vibration, temperature, and humidity loads experienced by the monitor at programmed intervals. The sensor was bolted to the chassis of the monitor to record its vibrations. The data recorder (shown in Fig. 2) was equipped with an internal triaxial piezoelectric accelerometer as well as an internal temperature and humidity sensor. The device can be used to record two types of data: dynamic (i.e., data that changes rapidly with time) and static (data that can be completely characterized by a single reading). In order to prevent any overlapping of dynamic and static data, memory is divided into two partitions: • Signal-triggered: This option allows the user to specify the minimum value of a parameter that should be

Sampling rate 2:56 ð2Þ

and Number of analysis lines ¼

Sample size ðN Þ 2:56

The number of analysis lines was set at 400 (the maximum), and the maximum frequency was set at 580 Hz (a sufficiently high number for road vibrations). With these, the sampling rate and sample size were set at 1500 samples/second and 1024, respectively. The corresponding time interval between two sample values (h ¼ 1=sampling rate) was 0.66 ms and the total record length (T ¼ Nh) was 683 ms. The signal and time triggers were set at ±2.0 g and 10 min, respectively. A low-pass filter (set at 40% of the sampling rate) was used to screen out all frequencies above the Nyquist frequency to prevent aliasing [13].

336


Fig. 3. Five-day sample of vibration data.

The data recorder was programmed to record vibration, temperature, and relative humidity. The sensor recorded data from 2:39 p.m., October 11, 2000 to 12:48 p.m., October 16, 2000. Federal Express picked up the monitor from College Park, MD at 6:59 p.m., October 11, 2000 and delivered it to San Pedro, CA at 12:45 p.m., October 12, 2000. On its return journey, the monitor was shipped from San Pedro at 6:24 p.m., October 14, 2000 and received in College Park at 12:44 p.m., October 16, 2000. All times are according to Eastern Standard Time (EST).

4. Analysis results Vibration signals can be analyzed in a number of domains (such as the time, frequency, amplitude, or power domains). The time domain mode involves using time as the independent variable to analyze transient signals. The analysis of a vibration is said to be a waveform analysis if it displays signal amplitude versus time. The frequency domain mode involves using frequency as the independent variable to analyze transient signals, such as those generated by rotating machinery. The analysis of a vibration is said to be in the frequency domain if it displays signal amplitude versus frequency. Fig. 3 shows the five-day sample of transportation data in which the root mean square value of the recorded acceleration (gRMS ) is plotted versus time. Since acceleration was measured as a signal-triggered quantity, the time axis is not linear. It only shows the in-

stances when the acceleration trigger of ±2.0g was exceeded. 1 Since the time plot of the acceleration does not indicate the frequencies present in the vibration, the usual variable for representing vibration is the power spectral density (PSD). The PSD describes the frequency composition of the vibration in terms of its mean square value over a frequency range. For sampled data (i.e., the type of data recorded in this paper), the PSD is calculated by the Cooley–Tukey method, which is based on computing the PSD via a Fast Fourier Transform (FFT) of the original sampled acceleration data. For a sequence of acceleration values hk sampled over a record length T , the Cooley–Tukey method defines the PSD function at any frequency f as [14]: Gðf Þ ¼

2h jXk j2 N

ð3Þ

where Xk are the FFT components of the N sampled acceleration values of amplitude hk averaged over the record length T . Fig. 4 shows the average power spectral density (PSD) plot of the acceleration data in Fig. 3 (from October 11 to 16, 2000). The three curves repre-

1

The signal trigger of ±2.0g refers to the magnitude of the actual acceleration, and not its root mean square value. In other words, Fig. 3 shows the root mean square value of the acceleration for those time intervals when the magnitude of the actual acceleration exceeded ±2.0g.


337

Fig. 4. Power spectral density curves for the vibration data in Fig. 3.

Fig. 5. Five-day sample of temperature data.

sent the PSD curves of the acceleration recorded by the triaxial accelerometer in the X , Y , and Z directions. The PSD curves are averaged over the recording interval T (683 ms). When compared to Fig. 1, Fig. 4 shows slightly higher values of PSD for the dominant frequency, which is possible because Fig. 4 also takes aircraft vibrations into account. 2

Fig. 5 shows the temperature and humidity profiles for the five days. Each data point is taken at an interval of 10 min. The sensor reports that the temperature during flight is the lowest, and the temperature in California is the highest.

5. Conclusions

2 According to MIL-STD-810F, aircraft vibrations are higher than highway (road) vibrations [12].

This paper presents a practical approach to characterize a commercial vibration and shock environment during transportation for physics-of-failure based reliability assessment. It provides the reader with a sample of

338


a complete transportation environment, including actual acceleration magnitudes induced by the vibration, temperature and humidity profiles, and power spectral density curves. Characterization of the real environment will help enable more accurate reliability predictions, more robust designs, and more realistic accelerated test planning.

[4]

[5] [6]

Acknowledgements The research for this paper was performed at the CALCE Electronic Products and Systems Center of the University of Maryland. The Center provides a knowledge and resource base to support the development of competitive electronic components, products and systems. The Center is supported by more than 100 electronic products and systems companies from all sectors, including telecommunications, computer, avionics, automotive, and military manufacturers. Special thanks is given to Mr. Michael Tierney of Measurements Inc. for his advice and support.

[7]

[8]

[9] [10]

[11]

References [1] Pecht M, Nash F. Predicting the reliability of electronic equipment. Proc IEEE 1994;82(7):992–1004. [2] Pecht M, Dasgupta A, Barker D, Leonard CT. The reliability physics approach to failure prediction modeling. Qual Reliab Eng Int 1990;6:267–73. [3] Pecht M. Reliability engineering in the 21st century––a focus on predicting the reliability of electronic products

[12]

[13] [14]

and systems. In: Proceedings of the 5th ICRMS 2001, Dalian, China, August 2001, vol. 1. p. 1–19, 28–31. Pecht M, Ramakrishnan A. Development and activities of the IEEE Reliability Standards Group. J Reliab Eng Assoc Jpn 2000;22(8):699–706. Dasgupta A, Pecht M. Failure mechanisms and damage models. IEEE Trans Reliab 1991;40(5):531–6. IEC Standard 60134. Rating systems for electronic tubes and valves and analogous semiconductor devices (Last reviewed in July 1994 by the IEC Technical Committee 39 on Semiconductors), 1961. Das D, Pendse N, Pecht M, Condra L, Wilkinson C. Understanding electronic part data sheets for parts selection and management. IEEE Circ Dev 2000;16(5):26– 34. Cluff KD, Robbins D, Edwards T, Barker D. Characterizing the commercial avionics thermal environment for field reliability assessment. J Inst Environ Sci 1997;40(4): 22–6. Collacott RA. Vibration monitoring and diagnosis. London: George Goodwin Ltd.; 1979. p. 15. Webster JG. The measurement, instrumentation, and sensors handbook. CRC Press/Springer-Verlag/IEEE Press; 1999. Carstens JR. Electrical sensors and transducers. New Jersey: Regents/Prentice Hall; 1993. p. 158–63. U.S. MIL-STD-810F. Department of Defense Test Method Standard for Environmental Engineering Considerations and Laboratory Tests. Version F, US Government Printing Office, January 2000. Dallas Instruments. SAVERe User’s Manual. Dallas, Texas, February 2000. Bendat SJ, Piersol AG. Random data: analysis and measurement procedures. New York: Wiley-Interscience; 1971.