PSI - Issue 5

S. Sahnoun et al. / Procedia Structural Integrity 5 (2017) 1267–1274 A.Saifi et al/ StructuralIntegrity Procedia 00 (2017) 000 – 000

1272

6

Fig 7: Superposition of the values obtained by simulation and those obtained by the proposed model

To evaluate the precision of our model, we have calculated the mean deviation given by the following relation: The i

n i  

(3)

n y y i sim 

mod

1

error



error of the model obtained is equal to 0.1286, which shows the aptitude of the proposed model to be used for the detection of the scale thickness in the steel water pipes. 4. Estimation of scale thickness by artificial neural networks.. In order to increase the accuracy of the estimation of scale thicknesses in water distribution systems. We used the technique of artificial neural networks. 4.1. Un neuron artificial An artificial neuron is inspired by the functioning of the biological neuron. It receives the information coming from the inputs (x i ) via the connections to which each of them is assigned a weight w ij weighting the information and also representative of the connection force (figure8).

Fig 8: Artificial neuron model

An artificial neuron consists of an addition unit and an activation function:

 The addition unit is a summation operator that develops a « potential post-synaptique » p j of neuron j equal to the weighted sum entries of the cell to which is added a constant term called bias of entry b j which can be considered as the weight of an input xi equal to 1:

n i 1 ji

j P b     j

(4)

w x

i

 The activation function is an operator of decision f (p j ) who calculates the State of the output y j f the neuron