پیش‌بینی ظرفیت باربری وابسته به زمان شمع‌های کوبشی در خاک چسبنده با استفاده از روش گروهی مدیریت داده‌ها

وابستگی ظرفیت باربری شمع‌ها به زمان، از مهم‌ترین مسائل در زمینه‌ی طراحی شمع‌هاست. عموماً ظرفیت باربری شمع‌های اجرا شده در خاک‌های چسبنده بعد از اتمام فرایند نصب و با گذشت زمان افزایش می‌یابند که آن را پدیده‌ی گیرش می‌نامند. در مطالعه‌ی حاضر، برای تخمین میزان گیرش و ارائه‌ی رابطه‌یی کابردی برای استفاده در مقاصد طراحی پی‌های عمیق، از روش گروهی مدیریت داده‌ها (g‌m‌d‌h) که یک الگوریتم هوشمند و داده‌محور است، استفاده شده است. بانک داده‌ی استفاده شده در پژوهش حاضر، شامل خصوصیاتی از شمع و خاک بوده است که به عنوان متغیرهای مستقل به سیستم معرفی شده‌اند. نتایج حاصل از تحلیل حساسیت مدل نشان می‌دهد که پارامترهایی، نظیر ظرفیت باربری اولیه و مقاومت برشی زهکشی نشده‌ی خاک با درصد مشارکت به ترتیب 37 و 22، تاثیر به‌سزایی در افزایش ظرفیت باربری وابسته به زمان شمع دارند. چندجمله‌یی حاصل از مدل‌سازی برای پیش‌بینی ظرفیت باربری وابسته به زمان در مطالعه‌ی حاضر، باضریب تعیین 0/76 می‌تواند به عنوان رابطه‌یی قابل قبول در فرایند طراحی شمع استفاده شود.

Prediction Of TimeDependent Bearing Capacity Of Pile Driven In Cohesive Soil Using Group Method Of Data Handling

Evaluating the ultimate bearing capacity of piles has been always an important concern for geotechnical engineers. Pile setup is a term which refers to an increase in bearing capacity of pile after a specific time. This increase is mainly considered relevant to the dissipation of excess pore water pressure created as a result of disturbance of the soil around the pile. Many researches have been centered on the investigation of pile setup and the factors influencing that. Results indicate that soil and pile properties can affect the occurrence and intensity of this phenomenon. The application of artificial intelligence such as artificial neural networks and evolutionary algorithms are considered as efficient and powerful methods for prediction and function finding purposes. Group Method of Data Handling is an intelligent approach that operates in a similar pattern to artificial neural networks. In this system, dual combinations of input variables are created in the form of KolmogorovGabor polynomials. Based on the evaluation criteria such as Root Mean Squared Error (RMSE) and determination coefficient (R^2), the polynomials with higher accuracy are selected and introduced to the next layer as inputs. This repetitive approach is used to reach the best polynomials predicting the target variable of the project. GMDH is a selforganized system in which the number of required layers and neurons are determined during the running process. In this paper, a dataset obtained from the literature review and contains information about 170 test piles derived in clay and mixed soil, is used in which the ultimate bearing capacity is considered as a target, while the other ones are independent variables. It is noticeable that to evaluate the efficiency of the ultimate model, data is randomly divided into training and testing data which the former includes 118 and the latter has used 52 data. The results of this study indicate that the initial bearing capacity of pile (End Of Driving) and undrained shear strength ( S_u) have a significant effect on the timedependent increase in bearing capacity of pile. The ultimate model obtained from the GMDH system with the RMSE of 742.85 and the (R^2) of 0.768 is an acceptable equation that can be used in the process of pile design.