ارايه‌ي مدل تخمين پاسخ غيركشسان سازه هاي يك درجه‌ي آزادي كاهنده ي معادل تحت ركوردهاي ثبت‌شده در خاك‌هاي سخت به وسيله ي شبكه هاي عصبي مصنوعي

An Artificial Neural Network Based Method for Estimation of the Inelastic Response of Single-Degree of Freedom (SDOF) Systems under Records of Rock.

با توجه به گسترش روزافزون مباحث طراحي براساس عملكرد در ايران نياز به رابطه ‌يي جهت تخمين پاسخ غيركشسان سازه هاست. در اين مطالعه نيز سعي بر آن است كه پاسخ غيركشسان سازه ها با ايجاد يك مجموعه داده ي مرجع بدون انجام تحليل بر روي سازه ها به‌دست آورده شود. براي اين منظور از شبكه‌هاي عصبي مصنوعي استفاده شده است. در اين نوشتار پاسخ سازه هاي يك درجه‌ي آزادي با ويژگي هاي مختلف و 3 نوع رفتار كاهنده تحت انواع ركورد هاي مختلف زلزله به‌دست آورده شده است. با توجه به پراكندگي نسبتاً زياد ركوردهاي موجود براي كمينه‌سازي مقدار خطا در تخمين پاسخ غيركشسان، نتايج مورد استفاده در مرحله‌ي آموزش شبكه‌هاي عصبي بر اساس دوره‌ي تناوب غالب ركوردهاي زلزله تقسيم‌بندي شده اند. و نهايتاً با توجه به شبكه هاي عصبي مصنوعي بهينه، پاسخ سازه هاي يك درجه‌ي آزادي با زوال سختي و مقاومت تحت انواع ركورد هاي مختلف زلزله پيش‌بيني شده است.

As a region with a high level of seismicity, Iran has been exposed to a prodigious number of destructive seismic events during its history, causing considerable loss of life and financial damage. Improvements in engineering knowledge about the nature of seismicity and its concomitant effects on structures have been behind the development of numerous analytical methods regarding both structural behavior and related damage induced by seismic events. Nowadays, concomitant with introducing the concept of performance based design methods and increasing the importance of using structural inelastic displacement parameters instead of forces, the need to apply non-linear rather than linear analysis has grown. In this regard, the non-linear static method has recently become increasingly popular for practical use. This method is simpler than dynamic methods and needs less computational time. However, this method is comprised of limits and problems. Due to the mentioned necessity, the substantial growth in application of performance based design methods in Iran requires equations which provide estimates for a structures’ inelastic response. Thus, the main objective of this paper is to develop a method to represent an estimate of the inelastic response for structures of single degree of freedom reductions. To this end, Artificial Neural Networks have been used, and responses for single degree of freedom structures with different characteristics and three types of reduction behavior have been represented under different kinds of seismic record. From the resulted outputs, 70% to 80% of data were used for training the neural network model and the remaining 20% to 30% were used for validation. Due to the relative high scattering of available records, and in order to minimize the error in estimating inelastic response, the outputs used in the training stage of developing the neural networks have been classified based on prevalent periods of seismic records. Finally, using the trained neural networks, software has been developed capable of predicting the response of single degree of freedom reduction structures under different seismic records.