تخمين نتايج آزمايشSPT بااستفاده از روش هاي احتمالاتي و شبكه عصبي در لايه هاي رسي( مطالعه موردي: لايه هاي رسي تبريز)

Estimation of SPT Results Using Probabilistic Method and Artificial Neural Network in clay layers (Case study: Tabriz Clayey layers

در اﯾﻦ ﻣﻄﺎﻟﻌﻪ ﺗﺨﻤﯿﻦ ﻧﺘﺎﯾﺞ آزﻣﺎﯾﺶ ﻧﻔﻮذ اﺳﺘﺎﻧﺪارد ﺑﻪ روشﻫﺎي اﺣﺘﻤﺎﻻﺗﯽ و ﻧﯿﺰ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ ﺑﺎ اﺳﺘﻔﺎده از ﺧﺼﻮﺻﯿﺎت ﺧﻤﯿﺮي و ﻓﯿﺰﯾﮑﯽ ﻻﯾﻪﻫﺎي رﺳﯽ و ﺑﺎ ﻣﻄﺎﻟﻌﻪ ﻣﻮردي ﺧﺎكﻫﺎي رﺳﯽ ﺷﻬﺮ ﺗﺒﺮﯾﺰ اﻧﺠﺎم ﺷﺪه اﺳﺖ. در اﯾﻦ ﺗﺤﻘﯿﻖ دو ﮔﻤﺎﻧﻪ ﻣﺎﺷﯿﻨﯽ ﺗﺤﻘﯿﻘﺎﺗﯽ ﺑﻪ ﻋﻤﻖ ﺗﺎ 8 ﻣﺘﺮي ﺣﻔﺎري ﺷﺪه و آزﻣﻮنﻫﺎي آزﻣﺎﯾﺸﮕﺎﻫﯽ ﺑﺮ روي ﻧﻤﻮﻧﻪﻫﺎي ﺗﻬﯿﻪ ﺷﺪه و ﻧﯿﺰ آزﻣﺎﯾﺶ ﺻﺤﺮاﯾﯽ ﻧﻔﻮذ اﺳﺘﺎﻧﺪارد اﻧﺠﺎم ﮔﺮﻓﺖ. ﺑﺎ اﺳﺘﻔﺎده از ﻧﺘﺎﯾﺞ آزﻣﺎﯾﺸﺎت و دادهﻫﺎي ﻣﻮﺟﻮد، ﺑﺎﻧﮏ اﻃﻼﻋﺎﺗﯽ ﻣﺸﺘﻤﻞ ﺑﺮ 112 ﺳﺮي از ﺧﺼﻮﺻﯿﺎت ﻻﯾﻪ ﻫﺎي رﺳﯽ ﻣﺤﺪوده ﺗﺤﻘﯿﻖ ﺗﻬﯿﻪ ﮔﺮدﯾﺪ. ﺑﻪ ﮐﻤﮏ روش ﺗﺤﻠﯿﻞ اﺣﺘﻤﺎﻻﺗﯽ ﺳﻪ ﺳﯿﮕﻤﺎ ﻣﺤﺪوده ﻫﺎي اﺣﺘﻤﺎﻻﺗﯽ ﺑﺮاي ﻫﺮﮐﺪام از ﻣﺘﻐﯿﺮﻫﺎي ﺗﺤﻘﯿﻖ ﺷﺎﻣﻞ ﻓﻌﺎﻟﯿﺖ، درﺻﺪ رﻃﻮﺑﺖ، ﺷﺎﺧﺺ اﺳﺘﺤﮑﺎم، درﺻﺪ ﻋﺒﻮري اﻟﮏ 200 و داﻣﻨﻪ ﺧﻤﯿﺮي ﺑﻪ ﻣﻨﻈﻮر ﺗﺨﻤﯿﻦ ﻧﻔﻮذ اﺳﺘﺎﻧﺪارد ﺗﺼﺤﯿﺢ ﺷﺪه ﭘﯿﺸﻨﻬﺎد ﮔﺮدﯾﺪ. ﺑﻬﺘﺮﯾﻦ راﺑﻄﻪ ﻣﺮﺑﻮط ﺑﻪ ﻧﻔﻮذ اﺳﺘﺎﻧﺪارد ﺑﺮاﺳﺎس ﺷﺎﺧﺺ اﺳﺘﺤﮑﺎم ﺑﺎ ﺷﯿﺐ ﻫﻤﺒﺴﺘﮕﯽ 1/015ﻣﯽ ﺑﺎﺷﺪ. ﻫﻤﭽﻨﯿﻦ ﻣﻨﺎﺳﺐ ﺗﺮﯾﻦ ﻧﺘﯿﺠﻪ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ ﻣﺮﺑﻮط ﺑﻪ دو ﻻﯾﻪ ﭘﻨﻬﺎن ﺑﺎ ﺗﻌﺪاد 10و 5 ﻧﺮون ﻣﺨﻔﯽ ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺎ ﺿﺮﯾﺐ ﺗﻌﯿﯿﻦ و ﺧﻄﺎي رﯾﺸﻪ ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺑﺮاﺑﺮ 0/805 و0/063 در ﻣﺮﺣﻠﻪ ﺗﺴﺖ ارزﯾﺎﺑﯽ ﮔﺮدﯾﺪ. ﻧﺘﺎﯾﺞ روش ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻧﺴﺒﺖ ﺑﻪ روش اﺣﺘﻤﺎﻻﺗﯽ ﺑﻪ ﺻﻮرت ﺎري ﻣﻘﺎﯾﺴﻪ ﮔﺮدﯾﺪه و ﻣﻨﺎﺳﺐﺗﺮ ﺑﻮده اﺳﺖ.

In this study, the estimation of the results of the standard penetration test with probabilistic methods and artificial neural network using the physical and plasticity properties of clay layers with case study of clayey soils of Tabriz city have been performed. In this research, two machine boreholes were drilled at depths of up to 8 meters and standard penetration tests and other tests were carried out to determine all of the plasticity and physical properties on the prepared specimens. By using the results of experiments as well as available data, a database of 112 series of clay layer properties were prepared. The probabilistic ranges for each of the research variables including activity, moisture content, consistency index, passage percentage sieve 200 and plasticity index were proposed for estimating the corrected standard penetration by using the three sigma probabilistic analyses. The best correlation with the standard penetration based on the consistency index with the correlation slope is 1.015. Also, the artificial neural networks were studied in different states and with the different number of hidden neurons with all physical and plasticity properties including 11 input variables. The best results of the artificial neural network related to the hidden double layer with 10 and 5 hidden neurons that the determination coefficient and the root mean square error were equal to 0.805 and 0.063 at the test stage, respectively. The results of the neural network method have been statistically compared and more appropriate than the probabilistic method.