This document discusses the history and limitations of using failure rate models like MIL-HDBK-217 to predict reliability. It notes that while these models were useful in the 1960s based on failure patterns at the time, they make unrealistic assumptions that are no longer valid for complex modern systems. Specifically, these models assume constant failure rates, ignore quality factors, and treat a system's reliability as simply the sum of its component reliabilities. The document concludes that comprehensive knowledge of failure mechanisms and interactions is now needed to accurately predict reliability.
2. History of Reliability
US Dept. Of Defence: Military systems => extreme failure rates
AGREE (1952):
50
•Reliability = integral part of development
19
•Derating
•Test @ high Stress @ 1000’s of hours
60
•MTBF & Statistics
19
70
19
80
19
90
19
00
20
2
AGREE: Advisory Group on Reliability of Electronic Equipment
3. History of Reliability
MIL-HDBK-217 (1962) started
50
19
•Failure rate is constant (λ)
60
19
•MTBF = 1/ λ
70
19
MIL-HDBK-217 cancelled
80
19
90
19
t. 2
no
00
: F-
-2 8
20
- 02
3
95
19
4. Predicting the failure rate, MIL-HDBK-217
Electrolytic capacitors:
λP = λbπ CV π Qπ E Failures / 10 hours
6
Microprocessors:
λP = (C1π T + C2π E )π Qπ L Failures / 10 hours
6
•
•
•
For the entire system:
i =n
1
λSYSTEM = ∑ λP ,i ⇒ MTBFSystem =
i =1 λSystem
4
5. Necessary assumptions for prediction:
•Constant failure rate
•1 Toyota for 7.000 hours = 7.000 Toyota’s for 1 hour
•A new car fails just as often as an old car that has run 300.000 km
•System = sum of its components
•No tolerance problems. No interface problems.
•SW quality doesn’t matter!
•1 line-of-code identical to 70.000 lines-of-code
•Mechanical quality doesn’t matter!
•However, you can use NPRD-95 (Non-electronic Parts Reliability Data), which is
much less refined than MIL-HDBK-217
•All system manufacturers have identical production quality
•Only 14 different environments (of which 11 are military)
•The authors of MIL-HDBK-217 know the quality of your product.
5
6. Basic belief for the use of MIL-HDBK-217
- And it’s many sisters: HRD-5, RDF-2000, Italtel, Telcordia/Bellcore …
•Product reliability is inherent in the components. When a
component fails, the cause should be found in the
component itself.
•This was more true in the 1960’ies when the failure
pattern was caused by many electronic component defects,
due to low production quality.
•Today, this viewpoint is obsolete.
Grundfos does not use this type of reliability prediction.
6
7. ”To meet any reliability objective requires comprehensive
knowledge of the interaction of failure mechanisms, failure
modes, the mission profile, and the design of the product.”
J1879, Handbook for robustness
validation of semiconductor devices
in automotive applications
7