PSI - Issue 19

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

137

8

analysing life situation measurement with a statistical target: measurement repetition of each usage for different users. Strength profile statistical distribution should be assessed by process monitoring and characterization tests [7] . The test factor depends on the confidence level associated to the reliability demonstration, the number of samples, and the statistical distribution law of the median-related parameter of the strength profile. This last parameter can be explained the sampling theory: from a batch to another, statistical properties, so as the median, will differ. Each statistical property is thus considered as a random variable governed by a statistical distribution law.

Fig. 5. Warranty and Test Factor for test specification

The application of the warranty factor and the test factor could in some cases lead to disqualified load level, because exceeding the failure mechanism boundary conditions. Demonstrating a reliability target for high cycle fatigue with a stress level leading to local plastic deformation when considering the stress – strain curve is not mechanically acceptable. In that case, the stress level must be lowered, which implies the increase of the number of cycles (accelerated life model) and then the test duration (but much less than to make a new test specification after a non representative series of tests, which also implies unexpected costs). 4. Real-Time analysis Since “Real - Time” often connotes data science, it is important to frame the term as an introduction to this section. With the development of technology, the new data collection and management ability leads to the “re” rise of data science. Processing data during tests should have lots of benefits but it is today at experimentation stage and needs further work to be deployed on test bench. “Real - Time analysis” could be defined as the ability to exploit an information as soon as it is available. For some appl ication, “Real - Time” is set to an observation window of about 10 µs (ultra-sonic inspection for instance), and for some other application, it can be set to a window of about 10 years (in-service fatigue failure analysis for instance). When monitoring a test bench to master the load history or to monitor health of components, the Real-Time window should be much smaller than the one needed to compute a test results (failure at a specific time) into a reliability assessment. This section will exclusively focus on this last application. For any test strategy involving more than one unique sample and leaded until the failure of the sample, progressive analysis of the test results will allow a progressive assessment of the system reliability: at test number X, reliability demonstrated is assessed to Y%, etc. This simple approach allows to stop the series of tests at strategic times defined by gain functions or convergence functions. Care is however needed: stopping a series of tests as soon as the reliability target is reached could lead to erroneous results (see Figure 6). The test program defined before the test realization is usually fully needed to the reliability demonstration, not partially. Stop criterion is efficient only if reliability estimates converge to a higher or lower level than the target. Higher level may convince to validate the technical solution sooner, lower level may convince to define an action plan without spending useless time and money.