EXPERIMENTAL STUDIES AND ANN MODELLING OF WIND LOADS ON LOW BUILDINGS

Abstract

Wind loading on low building has been an area of active investigation due to increasing public concern towards severe damage caused by windstorms in recent years. Design wind loading is mainly obtained from different codes, wherein the major source of information on which specifications are based is wind tunnel testing of scaled rigid models under simulated flow. Recent studies on full-scale/model-scale comparisons have brought out issues of flow simulations and measurement techniques with greater clarity, thus enabling improvements in existing information. The study reported in this thesis is an effort in that direction for a gable roof building. In carrying out such a study, it has been recognised that wind loads are significantly affected by building geometry, and the angle ofwind incidence. Besides a study of gable buildings in the 'stand-alone' situation, effect of interference by similar buildings has been studied. An effort has also been made in the present work to examine the efficiency of the use of Artificial Neural Network modelling for predicting the wind-induced pressures on the basis of data obtained from wind tunnel experiments. In the first phase of the experiments, wind tunnel study of gable building models of different roof slopes has been carried out in stand-alone position under simulated flow conditions. The building selected for the study is a hypothetical low gable roof building without openings. The length, width and eaves height of the buildings are 13.5 m, 7.0m and 5.1m respectively. Roof slope of the building has been varied from 10° to 35° with an increment of 5°. The Present study has been carried out on 1/100 scale models with the approach terrain classified as "open terrain with well scattered obstructions having heights between 1.5m to 10m" pertaining to category-2 of Indian Standard Code (IS 875 part 3, 1987). The mean flow profile is to conform to a power law index of 0.14. Different roughness devices have been used to develop appropriate turbulent flow for Atmospheric Surface Layer (ASL) simulation in the lower part (up to 60 cm), maintaining velocity profile and other important flow simulation parameters in the wind tunnel. Surface pressures from the various pressure taps on the model roof have been recorded with the help ofa tubing system with a flat frequency response upto 100 Hz, at a sampling frequency of 400 samples/sec/channel. 8192 samples of pressure data from each channel have been recorded, thus giving a record of approximately 20 seconds. The measured pressures are reduced to pressure coefficients based on a reference velocity, measured at the eaves height of the building. The pressure-time records have then been processed to give the values of mean, rms and peak pressures. Further a numerical method based on the probability distribution ofthe measured peak has been employed to give the values for the design pressure coefficients. It is well known that in practice a building is often surrounded by a number of other buildings and structures. Thus the flow fields of various buildings interfere with each other and create a wind field, which is different in comparison to that for an isolated building, and one which is far more complex. Several studies on interference effects on low buildings reflect that generalisation of the magnitude and patterns of pressures as affected by interference has not been possible. In the second phase of the experiments, interference studies on a building model of 20° roof slope have been carried out. Interference effects have been studied in two steps. In the first step, effects due to the presence of a single similar building have been in studied, whereas in the second step, interference effects due to a pair ofsimilar buildings have been studied. In the absence of a suitable analytical or mathematical model for the determination of wind loads on low buildings, with or without including the interference effects, Artificial Neural Network modelling has been applied in the present study as a means for predicting the same. Neural Network representations are capable of developing functional relationships from discrete values of input-output quantities obtained from computational approaches or experimental results. This property of generalisation makes it possible to train a network on representative set of input-output examples and get good results for new input without training the network on all possible input-output examples. Back-propagation learning algorithm has been used herein to predict the design pressure coefficients for different zones of the building roof in stand-alone as well as for interference cases. In the isolated model study, variation of mean, rms, peak and design pressure coefficients for different zones of building has been studied for different angles of wind incidence. Critical directions of wind incidence have been identified for different zones of the roof. Further, the variation of pressure coefficients with roof slope (irrespective of wind incidence) for each zone on the roof has been studied. Comparison of design pressure coefficients is made with existing Indian Standard Code (IS875, Part -3, 1987) and American Standard Code (ASCE-7, 1995/98). Further, ANN Modelling is applied to predict the design pressure coefficients for various zones on the roof. Results of interference study are presented in the form of 'Interference Factor' (IF) defined as : IV „ _ Response in Interference Configuration Response in Stand - Alone Configuration Interference Factor contours of mean, rms, peak and design pressure coefficients for all zones ofthe building roof with change oflocation ofa single similar building have been presented. This is followed by the IF contours of mean, rms, peak and design pressure coefficients for the case of a pair of similar buildings. Further, ANN predictions of design pressure coefficients for all zones of the roof, for different locations of interfering building (for single building and for a pair of building) are compared with the experimentally obtained values. The study has yielded significant results for the isolated building as well as for the interference conditions. These are summarised as follows : i. Oblique wind directions are critical for pressures over most areas of the roof, however wind directions parallel to ridge and perpendicular to ridge are also observed to be critical for some areas of the roof, ii. The maximum values (irrespective of the wind incidence angle) of the mean, rms, peak (negative) and design pressure coefficients for local pressure zones, decrease as the roof slope increases. The change is sharp from 10° to 25° roof slopes and gradual from 25°to 35° roof slopes, iii. The maximum values (irrespective of the wind incidence angle) of the mean, rms, peak (negative) and design pressure coefficients decrease as roof slope increases for zones (other than those in ii above) on windward side of the roof and increase as roof slope increases for leeward side of the roof, iv. Experimentally obtained values of design pressure coefficients are found to be of the same order as given in Indian Standard Code (1S875 Part-3, 1987) and in VI. Vll. vm. IX. American Standard Code (ASCE-7, 1995/1998), however some variations exist between the codal values and the experimental values. Both shielding and amplification of suctions is observed for different zones, for different positions of interfering building (s). In the case of interference with a single building, the maximum enhancement has been observed in the rms value of pressures and maximum reduction (due to shielding) occurs for the mean pressures. Maximum enhancement in rms pressures has been found to be 40% for zone 1and maximum shielding in mean pressures is found to be 44% for zone 6. (Refer figure over leaf) In the case of interference with a pair of buildings also, maximum enhancement has been observed for rms pressures and maximum shielding has been found for mean pressures. Maximum enhancement in rms pressures has been found to be 48% for zone 1and maximum shielding in mean pressures is found to be 34% for zone 6. Artificial Neural Network Modelling is seen to predict successfully, the pressure coefficient for any roof slope not covered by the experimental study, based on data from other roof slopes. The maximum error seen in this study is 7%. ANN modelling trained on the discrete interference results, can predict design pressure coefficients for different zones of the roof for a more generalised interference situations. The results have been found to be within 5% of the measured values. ANN modelling reduces the wind tunnel testing for interference studies to almost half. VI 5.1 <- V -> 7.0 END ELEVATION A to 1 2 5 4 7 3 6 8 V <- 7.0 PLAN •> All Dimensions are in cms Locations of Different Zones on Roof of Building model Used