Instrumentation & Measurement Magazine 25-7 - 51

Dimensions Weight
Table 1 - Comparative analysis of MFL detection based on dimensions weight
MFL
Techniques
MLP
90×150×300 mm, 4 kg
90×195×300 mm, 8 kg
190×280×300 mm, 12 kg
RBFN
Far_Z-CNN
MLP
RBFN
Far_Z-CNN
MLP
RBFN
Far_Z-CNN
Accuracy
85
88
95
89
94
96
89
94
98
Fully Connected Layer: Output of pooling layer is the input
of the fully connected layer. In this layer, every single node in
the first layer is connected to each available node in the second
layer as:
||||Φ   Φ || 2 tK S Sn
Ω Tf T ftt 

Pf f 
by  LL

n
12
nn   121/ 1|| ,
  f f L
2  (24)
d

P = Id as in pooling by sub-sampling. Pooling by averaging
*
n0
to satisfy:
n
   as a sign of difficulty in the
dd
operator commutation with the conversion operator Tt . Output-generating
atoms 
sup  01 1
  
Sn
n
 Ψ, ,MPn nn
1,
n
nN{}n
  
(25)
Let Ω  be an acceptable module-sequence,
let  be pooling factors, and consider that operators
dd PL: dd
ML L and 22
n : 22
  
translation operator Tt as:
  
 ΦΩΩ ∣
t
n Tf f      t
Φn
2 tK
1 n
SS

2 ,
,
n
  convert with L
f f L  (26)
2 dd
First and second convolutional layers: consist of kernel filters
of 64 features with 3×3 size. As signal passes through first and
second convolution, it has dimension of 224×224×64. Output is
then sent to the max-pooling layer with a stride of 2.
Third and fourth convolutional layers: consist of 124 3×3 kernel
filters. Output will be reduced to 56×56×128 with max-pooling
layer of stride 2.
Fifth, sixth and seventh layers: contain 3×3 kernel size. This
layer has a max pooling layer with stride 2.
Eighth and thirteen layers: contain 3×3 kernel size. It has a maxpooling
layer of stride 1. This layer contains 512 kernel filters.
Fourteenth and fifteenth layers: consist of Hidden layers
which are fully connected and have 4096 units, followed by a
softmax output layer of 1000 units (sixteenth layer).
October 2022
Fig. 1. Comparative analysis of MFL detection based on dimensions weight of
90×150×300 mm, 4 kg.
IEEE Instrumentation & Measurement Magazine
51
Reliability
65
71
75
71
75
79
74
79
81
Detection Rate
81
85
86
75
79
83
79
81
85
Magnitude
of MFL
83
85
89
77
81
85
76
78
83
Performance Analysis
In simulations, a 6×19 steel wire rope with a diameter of 24 mm
and six strands (every strand containing 19 separate wires)
was chosen. The left ray is used to make the six strands.
Table 1, Fig. 1, Fig. 2 and Fig. 3 show comparative analysis
of MFL detection based on dimensions weight. Here, the dimension
weight has been calculated based on the variation of
90×150×300 mm, 4 kg, 90×195×300 mm, 8 kg, and 190×280×300
mm, 12 kg. These dimensions are calculated based on the wire
rope dimensions. The parametric analysis is carried out in
terms of accuracy, reliability, MFL detection rate and magnitude
of MFL. Existing techniques that are compared are the
MLP, the RBFN, and the proposed technique Far_Z-CNN.
Here, the proposed technique obtained optimal results in
terms of all parameters.
Table 2, Fig. 4 and Fig. 5 show comparative analysis of MFL
detection based on the wire rope diameter. Here, the wire
rope diameter has been calculated based on the variation of
16×19 mm, 20-49 mm. The parametric analysis is carried out
in terms of accuracy, reliability, MFL detection rate and magnitude
of MFL. Existing techniques compared are the MLP,
the RBFN and the proposed technique Far_Z-CNN. Here, the
Instrumentation & Measurement Magazine 25-7

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-7
