PSI - Issue 19

Xavier Hermite et al. / Procedia Structural Integrity 19 (2019) 130–139 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

134

5

2.3. Analysis of the experience feedback

When a technical solution is close to another one, whose most likely failure modes/mechanisms are known, the tests specification should include this knowledge and focus on these weaknesses. In the same way, the re-use of reliability proven solution might reduce a lot the tests need, especially if the service conditions are identical, since a single reliability demonstration would be needed. Experience feedback analysis results is very dependant of the quality of the data to process. For reliability related database, key features are: - Failure and survival events, and associated service duration (mandatory): ▪ Failure events concern every functional anomaly detection in service or during maintenance, ▪ Survival events concern every product without observed degradation nor failure at a specific time, as after a preventive inspection, - Failure mode and failure mechanism (mandatory): ▪ Nice to have, even if deeper data analysis may help to classify the failure into early, random or wear-out failure, - Causes (optional): ▪ Nice to have to focus on specific variables for failure reproduction, - Usage history (optional): ▪ Nice to have to focus on specific variables for failure reproduction, and to precise the mission profile of the product and to make reliability estimates more accurate, ▪ Usage history concerns climatic conditions and mechanical loading, ▪ Extremely difficult to have access to such data since embedded sensors and recorder are needed. As first step of analysis, failure mode and mechanism should be studied with the Pareto method to identify the most likely in-service phenomenon. Failure and survival events associated to the main failure modes/mechanisms should then be studied for reliability estimates. Since reliability is very dependant to the strength of the product and to its load conditions, estimates should often not be used “as - is”, and deviation between strength - and stress-related parameters must be studied. Strength-related parameters concern design, material, treatments, and manufacturing process. To consider similarities between two technical solutions, only the design should be different, and not so significantly in so that numerical computation should be enough to establish transfer functions. Stress-related parameters concern environment, usage and users of the product. If the database allows samples comparison to identify the influence factors on reliability estimates, transfer functions from a mission profile to another could be computed. If not, the service conditions of the similar product should not be significantly different to use the experience feedback. Reliability estimates are usually performed using the Weibull distribution because of the physical interpretation of its form parameter  . Depending on the quality of the data to process, the method to be used for Weibull parameters estimates can vary (frequency method, rank regression, maximum likelihood, etc., see Figure 4) [3, 4] . ( ) = (− ( − ) ) - t is the lifetime, assumed to evolve following a Weibull distribution, - R(t) is the probability of survival (reliability) at t time under Weibull distribution hypothesis, -  is the Weibull form parameter, which has mechanical interpretation: ▪  < 1: early failure, ▪  ≈ 1: random failure, ▪  > 1: wear-out failure, -  is the Weibull location parameter, usually set to 0 unless mechanical interpretation can explain the impossibility to observe a failure under that lifetime,