PSI - Issue 13

Mohammad Reza Khosravani et al. / Procedia Structural Integrity 13 (2018) 168–173 Author name / Structural Integrity Procedia 00 (2018) 000–000

172

5

4. Results and evaluation

In terms of system evaluation by an worked example, we consider our experimental results published in [30, 31] which show failure load and type of failure in a particular sa ndwich T-joint. Here, for the first evaluation in this study, we employed our CBR system to predict failure load and type of failure of our worked example as an incoming case in comparison with other 25 cases. We have calculated the res ults based on the similar tests classified as positive (TP: 4 cases) or negative (FN:2 cases) and for non-similar te sts classified as positive (FP: 1 case) or negative (TN:18 cases). Therefore, the accuracy of the proposed system for our experimental tests is 88%. The results of this use case is depicted in Fig. 3. Our implemented system can be considered as a NDE and utilize for high risk sandwich joints whose failure may lead to significant damage on the structure . It is expected this system predict failure of sandwich joints and provide a substantial risk reduction.

Fig. 3. The GUI example of the proposed system.

5. Conclusions and future work

Sandwich-structured composite joints experiences various loadings and environmental conditions during their ser vice life. The static and dynamic loading regimes and exposure to the high and low temperature can lead to the mechanical fractures. In this research, we developed an intelligent system based on a CBR approach to predict frac ture in adhesively bonded sandwich composite joints. To this aim, experimental results of tests on various sandwich joints are saved in case base of the system with details. In performance of a adhesively sandwich joint various param eters such as design of joint, utilized material, configurat ion, type of loading, adhesive thickness, length of the joint and adhesive properties are crucial. All of these parameter s are considered as essential features and their significant weights are calculated. Moreover, the case base of the syste m is enriched by several simulations finding which con ducted numerical simulations on di ff erent sandwich joints. We randomly hold-out 10% of cases to act as a test-set of our above-mentioned experiments and 90% of cases as the case base for prediction and recommendation. The overall accuracy of the proposed system is 71%. The prediction of failure loads and type of failure in the adh esively bonded sandwich composites play a significant role in improvement of the performance in these joints. Using the proposed intelligent system shows that estimation of mechanical behavior and prediction of failure load of adhesively bonded sandwich joints are highly easier and significantly faster than experimental setup or finite eleme nt simulation. Moreover, the proposed system can be used not only for the conditions which are described in the previous investigations, but also for the whole spectrum of mechanical and environmental conditions that might be appear during the service life of the sandwich composite joints. A further work on implementation of an intelligent system to predict fracture in various composite joints is planned.