كليدواژه :
برش پانچ , دال دوطرفه , الگوريتم ژنتيك , برنامه نويسي ژنتيك , هوش مصنوعي
چكيده فارسي :
دال هاي دو طرفه بتن آرمه يكي از سيستم هاي مرسوم سازه اي مي باشند. مزاياي اين دال ها باعث كاربرد زياد آنها در صنعت ساختمان شده است. ولي اين سيستم ها با مشكلاتي نظير برش پانچ مواجه هستند. روابط موجود براي پيش بيني برش پانچ بر اساس نتايج آماري آزمايش هاي موجود در تحقيقات گذشته بدست آمده اند. با اين حال اين روابط تقريبي بوده و داراي خطاي بالا مي باشند. هدف اصلي اين مقاله معرفي روشي قابل اعتماد و كاربردي براي محاسبه برش پانچ براي دال هاي نازك و ضخيم با استفاده از هوش مصنوعي است. براي اين كار از برنامه نويسي ژنتيك و برنامه ريزي جغرافياي زيستي براي پيدار كردن رابطه بين ظرفيت برش پانچ و پارامترهاي موثر بر آن استفاده شده است. ابتدا 267 داده آزمايشگاهي موجود جمع آوري شده است. سپس با استفاده از روش هاي مذكور رابطه اي براي پيش بيني مقاومت برش پانچ ارايه شده است. نتايج نشان مي دهد كه روش هاي مبتني بر هوش مصنوعي قادرند با خطاي متوسط كمتر از 2% در مقابل خطاي 14 الي 28 درصدي روابط سنتي آيين نامه ها مقاومت برش پانچ را پيش بيني كند.
چكيده لاتين :
Two-way slabs are one of the common structural systems. The benefits of such systems have led to extensive use of them in building construction. However, these systems are prone to pushing shear problem which causes sudden failure. There are lots of equations to predict punching shear of slabs. The main proportion of the existing equations are based on statistical results from previous experimental studies. However, these equations are approximate and have large errors. Therefore, more exact and reliable equations that can estimate punching shear capacity are desirable. The aim of this study is to propose an applicable method to predict punching shear in thin and thick slabs using artificial intelligence. For this reason Genetic Programming (GP) and Biogeography-Based Programming (BBP) are employed to find a relationship between punching shear and the corresponding effective parameters. GP that is inspired by natural genetic process, searches for an optimum population among the various probable ones. Two main operations of GP are crossover and mutation which make it possible to form new generations with better finesses. Unlike the GP, BBP is a Biogeography-Based Optimization (BBO) technique which is inspired by the geographical distribution in an ecosystem. BBP employs principles of biogeography to create computer programs. First, 267 experimental data is collected from the past studies. Next, using the aforementioned algorithms, a relationship to predict punching shear is proposed. To evaluate the error of prediction, several error functions including RMSE, MAE, MAPE, R, and OBJ are utilized. Matlab software is used to build the models of prediction. 10 different models are built and the one with the minimum error is selected. Based on the results, GP3 and BBP9 models could reach the best fitness. These models contain 3 sub-trees that use operators of plus, minus, multiplication, division, ln, sin, power 2, power 5 power 0.5, power 0.33, power 0.2, and power 0.25. Overall, the final tree includes several variables and integers, the variables are inputs of column dimension, effective depth, rebar ratio, compressive strength of concrete, and yielding strength of the rebars, and the output of punching shear capacity. The results of modeling are compared with recommended values of the ACI318 and EC2 codes. Comparison shows that code equations are scattered and therefore are not very reliable. Maximum error for both model and code equations occurs when the yielding strength of the rebars is low. Minimum estimation is related to GP and ACI codes with the ratio of 0.485 and 0.52, respectively which is due to very low thickness of the slab (41 to 55 mm). The maximum estimated shear belongs to ACI code in which the estimated value is two times the real one. Also, standard deviation of ACI values is about two times the others. Among the code equations, EC2 values yield more accurate results. However, GP and BBP models give much less mean error. Also, standard deviation of these methods is less than code values. In total, results show that the methods based on artificial intelligence are able to estimate pushing shear with around 2% error, compared to existing code equations which give 14-28% error.