پارامترهاي بهينه ميراگر جرمي تنظيم شده براي سازه‌هاي بلند به كمك شبكه‌هاي‌عصبي مصنوعي

Optimal Parameters of Tuned Mass Damper for Tall Buildings by neural networks

ميراگرهاي جرمي تنظيم شده، TMD نوعي از جاذب‌هاي انرژي هستند كه اگر پارامترهاي فركانس و ميرايي آن‌ها به‌خوبي تنظيم شوند مي‌توانند ارتعاشات سازه‌ها را كاهش دهند. براي شناسايي اين پارامترها روابط تحليلي و تجربي بسياري ارائه شده است، كه هر يك از اين روابط با ساده ‌سازي‌هايي در سازه و بارگذاري آن بدست آمده است. در اين مقاله از شيوه‌اي جديد براي بدست آوردن اين پارامتر‌ها استفاده شده است. ابتدا سازه‌هايي با تعداد طبقات مختلف (8 تا 80 طبقه) در محيط كد نويسي متلب ايجاد شده‌اند. پس از دست‌يابي به ماتريس‌هاي سختي، ميرايي و جرم اين سازه‌ها، برنامه وارد محيط سيمولينك مي‌شود، و سازه همراه با TMD در محدوده تقريبي پارامترهاي بهينه، تحت زلزله‌هاي مختلف قرار مي‌گيرد. در نهايت پارامتر‌هاي مناسب براي بيشترين كاهش در جابجايي سازه، شناسايي مي‌شوند. سپس با داشتن آمار كامل از سازه‌ها با ارتفاع‌هاي مختلف، كار آموزش شبكه عصبي آغاز مي‌شود. به اين ترتيب شبكه عصبي مورد نظر پس از آموزش ديدن مي‌تواند با دريافت ورودي‌هايي همچون فركانس و نسبت جرمي TMD، مقاديري همچون فركانس و ميرايي TMD را به‌عنوان خروجي ارائه كند. با استفاده از اين روش مشاهده شده است كه خطاهاي موجود در فركانس و ميرايي بهينه TMD، نسبت به روابط تجربي دن‌‌هارتوگ كاهش قابل ملاحظه‌اي داشته است. بنابراين استفاده از نتايج اين شيوه مي‌تواند به عنوان پارامترهاي بهينه TMD، با اطمينان بيشتري مورد استفاده قرار گيرد. در پايان، اين مقادير با روابط ارائه شده توسط دن هارتوگ روي يك ساختمان بلند مرتبه مقايسه شده است.

There are a variety of tools like vibration absorbers to reduce the vibration of structures by dissipating imposed energy. Tuned mass Damper is a vibration system with a mass and spring, that usually installed on the top of a structure. Tuned mass dampers are kind of absorbers that could reduce vibrations if the parameters of frequency and damping are well tuned. There are many analytical and empirical relations to identify these parameters obtained by structure simplification and loading. The paper demonstrates that neural networks can be used effectively for the identification optimal parameters of tuned mass dampers.In this paper, a new method is proposed to determine these parameters. For this purpose, several buildings with a number of different stories (8 to 80) are first created in MATLAB code. After obtaining the stiffness, damping and mass matrix of the structures, the program enters the Simulink environment and then structure with TMD tuned in the approximate range of optimal parameters is analyzed under different earthquakes. These ground motion records, including two near field and two far field records that suggested by the International Association of Structural control are used. To increase the number of data used to train artificial neural network and reduce uncertainty, in addition to the records mentioned, four other earthquake was considered. The TMD Mass, damping coefficient and frequencies is assumed as the design variables of the controller; and the objective is set as the reduction of the maximum displacement of the building. In the end, the parameters for the maximum reduction in lateral displacement of structures are identified. Then neural network training begins to have a complete database of structures with different stories. The number of hidden layer neural network are ten, and the number of layer's output are two considered. To train the neural network of Six hundred and fortyfour date has been used. Of these numbers, the Seventy percents are share of training data, and fifteen percents are share of Validation data, and fifteen percents are share of test data. Accordingly, the neural network after training can evaluate frequency and damping ratio of TMD based on input such as frequency and TMD mass ratio. Using this method, it has been observed that errors in the frequency and damping optimum TMD have been reduced particularly compared to empirical relations of Den Hartog. This reduces the errors caused by factors not considered in relation to the various reasons. For example, these factors maybe not yet identify, or because of the complexity of behavior have been simplified. However, using a neural network without fully understanding the parameters influencing the behavior of TMD exists, these factors indirectly to predict the behavior of TMD considered. So the results of this method can be used to optimize the parameters of TMD, to be used with greater confidence. Finally, these values are compared with those provided by Den Hartog in a highrise building. The results showed that for the building, parameters provided by neural network approximately 1 percent of the actual value of the optimal values is different.