Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 10, Number 8 (2015)

© Research India Publications ::: http://www.ripublication.com

IMPLEMENTATION OF ARTIFICIAL NEURAL NETWORK IN ANALYSING


SUPPLY CHAIN FOR SINGLE RETAILER AND MULTIPLE SUPPLIERS

J.Ralph Leeben1*, M.Saravanan2, and S.Purushothaman3


1
Assistant professor, Department Of Mechanical Engineering, Sivaji College of Engineering and
Technology, Manivila, KanyaKumari Dist-629170, ralphsam123@gmail.com
2
Principal, CR College Of Engineering, India - 625 301drmsaravanan@yahoo.com
3
Professor, Department of Mechanical Engineering, PET Engineering College, India-627117
3
dr.s.purushothaman@gmail.com

Abstract. This paper presents implementation of back propagation algorithm for analyzing supply
chain for single retailer and multiple supplier. Simulation of results is done using RISK software.
The output data of the RISK software has been used to train the back propagation algorithm. The
BPA network performs to more than 95% at an average in analyzing the supply chain.

Keywords: Supply chain, back propagation algorithm, neural network, Retailer, multiple supplier

Introduction

Supply chain is a concept describing a smooth flow starting from supplier to manufacturer to
consumer. The supply chain management involves in making a smooth flow of material /
information from one point to another point.Production, scheduling, inventory control and delivery
plans are parts of supply chain management (SCM). These facets of supply chain management have
to be measured and controlled in order to be competitive. Competition is now taking place between
separate supply chains rather than between individual organizations [7]. A supply chain is a network
of facilities that procure raw materials, transformthem into intermediate goods and then final
products, and delivers the products tocustomers through a distribution system [5,6].
When the production capacity is fixed and lead times result, thisreplenishment rule enables to
smooth the manufacturer's production, resulting in shorterorder-to-delivery times and more
balanced, peak shaving production schedules whichare beneficial for the manufacturer. Besides the
benefits realized through a smootherplanning, the manufacturer also realizes cost savings on its own
raw materials and/or component inventories [1].
In developing a replenishment rule one has to consider the impact on the inventory variance as well.
The manufacturer does benefit from smooth production, but dampening variability in orders may
have a negative impact on the retailer's customer service due to inventory variance increases [3,4]
quantify the variance of the net stock and compute the requiredsafety stock as a function of the
smoothing intensity. Their main conclusion is that when customer demand is i.i.d., order smoothing
comes at a price - in order to guarantee thesame fill rate, more investment in safety stock is
required. As a consequence, retailers, driven by the goal of reducing inventory (holding and
shortage/backlog) costs, prefer touse replenishment policies that chase demand rather than dampen
consumer demand variability.

Impact on customer service and safety stock


When demand is probabilistic, there is a definite chance of not being able to satisfysome of the
demand directly from stock. Therefore, a buffer or safety stock is requiredto meet unexpected
fluctuations in demand. The goal is to reduce inventory withoutdiminishing the level of service
provided to customers. When the retailer faces (andsatisfies) a variable customer demand, but
replenishes through a smooth order pattern,more safety stock is required to buffer the difference
between usage and supply. Areduction of order variations comes with the cost of an increase in
inventory variability [2].

6138
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 10, Number 8 (2015)
© Research India Publications ::: http://www.ripublication.com

Back propagation algorithm


A new improvement of Back Propagation Neural Network learning algorithms with
Adaptive gain has been done already. The BPA uses the steepest-descent method to reach a global
minimum. The number of layers and number of nodes in the hidden layers are decided. The
connections between nodes are initialized with random weights. A pattern from the training set is
presented in the input layer of the network and the error at the output layer is calculated. The error is
propagated backwards towards the input layer and the weights are updated. This procedure is
repeated for all the training patterns. At the end of each iteration, test patterns are presented to ANN
and the classification performance of ANN is evaluated. Further training of ANN is continued till
the desired classification performance is reached.
Steps involved in training BPA
Forward propagation [8.9]:
Step 1: The weights of the ANN are initialized.
Step 2: The inputs and outputs of a supply chain training patterns are presented to the BPA
network. The output of each node in the successive layers is calculated using equation (1)
o(output of a node) = 1/(1+exp(- wij xi)) (1)
Step 3: The error of a pattern is calculated using equation (2).
E(p) = (1/2) ∑(d(p) – o(p))2 (2)
Reverse propagación (Weight updation)
Step 4: The error for the nodes in the output layer is calculated using equation (3).
δ(output layer) = o(1-o)(d-o) (3)
Step 5: The weights between output layer and hidden layer are updated using equation (4).
W(n+1) = W(n) + (output layer) o(hidden layer) (4)
Step 6: The error for the nodes in the hidden layer is calculated using equation (5)
(Hidden layer) = o(1-o) (output layer)W(updated weights between hidden and output layer) (5)
Step 7: The weights between hidden layer(s) and input layer are updated using equation (6).
W(n+1) = W(n) + (hidden layer) o(input layer) (6)
The above steps complete one weight updation. Second pattern is presented and the above steps are
followed for the second weight updation. When all the training patterns are presented, a cycle of
iteration or epoch is completed. The errors of all the training patterns are calculated using equation
(7).
E(MSE) = E(p) (7)

Experimentation
This experiment illustrates the bullwhip effect in a multi-tier setting. A retailer experiences periodic
customer demands and orders from a manufacturer. The manufacturer then places orders from a tier
1 supplier, the tier 1 supplier places orders from a tier 2 supplier, the tier 2 supplier places orders
from a tier 3 supplier, and the tier 3 supplier places orders from another supplier not shown
explicitly in the model. The variability in orders tends to magnify upstream. The manufacturer’s
orders tend to vary more than customer demands, the tier 1 supplier’s orders tend to vary more than
the manufacturer’s, and so on.
The following assumptions.
1. Customer demands are independent and normally distributed with constant mean 1000 and
constant standard deviation 200.
2. Each member of the supply chain forecasts its future demands with a moving average process,
using the number of periods on an average. If the moving average=4, the manufacturer uses the
average of the 4 most recent customer demands as its forecast, the tier 1 supplier uses the average of
the manufacturer’s 4 most recent orders as its forecast, and so on.
3. Each member of the supply chain orders enough each period to make its final inventory equal to
an amplification factor (1 or greater) times its forecast. If the manufacturer’s current forecast, based
on the most recent 4 customer demands, is 1030, and its amplification factor is 1.2, it will order
enough from the tier 1 supplier so that its final inventory is 20% more than 1030. For simplicity, the
6139
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 10, Number 8 (2015)
© Research India Publications ::: http://www.ripublication.com

model assumes that orders arrive as soon as they are placed and that there is always sufficient
capacity in the system (no backorders).
Let F is a member’s final inventory,
S as its starting inventory,
O as its order,
D as its downstream demand,
MA as its moving average forecast, and
f as its amplification factor.
Then the model forces F = f*MA (8)
However, the inventory balance equation is F = S+O-D.
Solving for O gives,
O = f*MA-S+D (9)
If this is negative, O is replaced by 0.
The amplification factors are not unrealistic. In real supply chains, upstream suppliers often
magnify their forecasts, and hence order more than is really necessary, to be on the “safe” side. To
see the effect of these factors, the model is set up to run 5 simulations with various combinations of
the factors. The outputs are ratios of standard deviations of order sizes to the standard deviation of
customer demands. When the simulations is run, the amplification factors greater than 1 cause
larger variability ratios upstream. However, the amplification factors are not the only. The output
ratios average is above 100% even when the amplification factors are all equal to 1.
Table 1 Inputs for simulation
Parameters of normal demand distribution
Mean 1000
StdDev 200

Moving average span 4

Table 2 Simulation index


Sim index M factor T1 factor T2 factor T3 factor
1 1.00 1.00 1.00 1.00
2 1.50 1.50 1.50 1.50
3 2.00 2.00 2.00 2.00
4 1.00 1.33 1.67 2.00
5 2.00 1.67 1.33 1.00

Table 3 Inventory simulation


Retailer (R) Manufacturer (M) Tier 1 supplier (T1) Tier 2 supplier (T2) Tier 3 supplier (T3)
Amplification factors 1.00 1.00 1.00 1.00
Period Demand Inv start Order Inv final Inv start Order Inv final Inv start Order Inv final Inv start Order Inv final
1 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
2 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
3 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
4 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
5 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
6 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
7 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
8 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
9 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000
10 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000

Figures 1 to 4 present the relationship among Suppliers and retailer at different percentages. The
vertical black lines can be widened according to the required amount of results required among
supplier and retailer.

6140
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 10, Number 8 (2015)
© Research India Publications ::: http://www.ripublication.com

Results And Discussions

Fig. 1 Manufacture to Retailer

Fig. 2 TIER-1 Supplier to Retailer

Fig. 3 TIER-2 Supplier to Retailer

Fig. 4 TIER-3 Supplier to Retailer

Conclusions
Supply chain system has been simulated to obtain the inventory data. These data has been
used as training patterns for the BPA network. The training time of the network increases as the
number of training pattern increases. The accuracy of estimation of continuity in supply chain by
BPA is 87.5%.
6141
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 10, Number 8 (2015)
© Research India Publications ::: http://www.ripublication.com

References
[1] A. Balakrishnan,. J.Geunes, and M.Pangburn, Coordinating supply chains bycontrolling
upstream variability propagation. Manufacturing & Service OperationsManagement, 6(2), (2004)
163-183.
[2] Bertrand, J.W.M., Balancing production level variations and inventory variationsin complex
production systems. International Journal of Production Research, 24(5), (1986) 1059-1074.
[3] Disney, S.M. and Towill, D.R., On the bullwhip and inventory variance producedby an ordering
policy. Omega, 31,(2003) 157-167.
[4] S.M.Disney, I.Farasyn, , M.R.Lambrecht, D.R. Towill and W.Van de Velde, Taming the
bullwhip whilst watching customer service in a single supply chain echelon.European Journal of
Operational Research 173, (2006) 151-172.
[5] J.G.Kim, , D.C. Chatfield, T.P. Harrison, and J.C. Hayya, Quantifying thebullwhip effect in a
supply chain with stochastic lead time. European Journal ofOperational Research, 173, (2006) 617-
636.
[6] H.L. Lee, V.Padmanabhan, and S.Whang, Information distortion in a supplychain: the bullwhip
effect. Management Science, 43(4), (1997b) 546-558.
[7] Wieland, Andreas; Handfield, B. Robert, The Socially Responsible Supply Chain: An
Imperative for Global Corporations, Supply Chain Management Review.17(5) ,(2013).
[8] S.Purushothaman, Y.G.Srinivasa, (1998), A procedure for training artificial neural network with
the application of tool wear monitoring, International Journal of Production Research, 36 (3), 635-
651, (1998).
[9] S.Purushothaman,Y.G.Srinivasa, (1994), A back-propagation algorithm applied to tool wear
monitoring, International Journal of M/C Tools & Manufacturing, 34(5), 625-631, (1994).

6142

You might also like