A VIRTUAL SENSOR FOR

PREDICTING DIESEL ENGINE

NOX EMISSIONS

SHUBHAM KUMAR PARJAPATI

SG-14941
MECHANICAL ENGINEERING(8TH
SEMESTER)
 Introduction

 Virtual sensing enables more direct and robust emission control

allowing.
 Artificial neural network (ANN)-based engine modeling offers
the potential for a multidimensional, adaptive, learning control
system.
 ANN as a predictive tool for multi-cylinder diesel engine NOx
emissions.
 The neural network system is able to predict accurately real-time
engine exhaust emissions using readily measured engine
parameters.
 Artificial Neural Network

 The ability to learn from experience in order to improve their

performance.
 A powerful technique to solve many real world problems.
 A method of computing, based on the interaction of multiple
connected processing elements.
 Ability to deal with incomplete information.
 BIOLOGICAL NEURON MODAL

 Four parts of a typical nerve cell:-

 DENDRITES: Accepts the inputs

 SOMA: Process the inputs
 AXON: Turns the processed inputs
into outputs.
 SYNAPSES: The electrochemical
contact between the
neurons.
 ARTIFICIAL NEURON MODEL

 Input to the network are represented

by the mathematical symbol, Xn
 Each of these inputs are multiplied by
a connection weight, wn
sum = w1x1+.............+wnxn
 These products are simply summed, fed
through The transfer function to
generate a result and then output.
 Activation Functions

 Controls when unit is ‘active’ or ‘inactive’

 Threshold function outputs 1 when input is positive and 0
otherwise
 Sigmoid function = 1/ (1+ exp(-x))
 NEURAL NETWORK ARCHITECTURES

 Single layer of neurons:-

consist of input layer and output layer.

 Multiple layer of neurons:-

consist of input layer, hidden layer,
and output layer.
 CLASSIFICATION BASED ON LEARNING
METHOD

 Supervised
 Unsupervised
 Reinforcement

Supervised learning
 Each training pattern: input + desired output
 At each presentation: adapt weights
 After many epochs convergence to a local minimum
Unsupervised Learning

 No help from the outside

 No training data, no information available on the desired output
 Learning by doing
 Used to pick out structure in the input:
 Clustering
 Reduction of dimensionality compression
Reinforcement learning

 Teacher: training data

 The teacher scores the performance of the
training examples
 Use performance score to shuffle weights
‘randomly’
 Relatively slow learning due to
‘randomness’
 EXPERIMENTAL SETUP

 Test engine

 Engine specifications

Engine type BS-III, DI, 4-stroke turbocharged, water cooled, diesel

engine
Engine 5.9 x 10-3 m3
displacement
Firing order 1–5–3–6–2–4
No. of cylinders 6
Maximum Horse 180 hp @ 2500 rpm
Power
Peak torque 680 N m @ 1400 rpm
Fuel injection Distributor injection pump
 CONT…

 Engine dynamometer
CONT….

 Emission analysers
CONT…..

 Fuel flow meter : Fuel consumption measurements were

performed for all fuel types
 ANN MODEL TRAINING

 Input X : Speed and Fuel

Flow Rate
 Target Y : NOx and Torque
ANN MODEL PREDICTION
 RESULTS AND DISCUSSION

 Modeling with the ANN

 70% of the total data used as a training data.
 Rest 30% of the total data used in testing and validation.
 Taking the 50 neurons in the hidden layer.
 Correlation coefficient was 0.92
CONT…..

 Neural network training regression

CONT….

 Comparison between experimental and Predicted NOx

 CONCLUSION

 Real time estimation of NOx values.

 Time and cost efficient
 Helpful for the new design and development for
Engine
 Long experimental processor does not required
 REFERENCES

 Fischer, M.I.C.H.A.E.L. 2013. Transient NOx Estimation

using Artificial Neural Networks. Advances in Automotive
Control. 7(IFAC AAC 2013), pp. 4-7
 Mohammad hassani et al... 2015. Prediction and reduction of
diesel engine emissions using a combined ANN–ACO method.
Applied Soft Computing. 34(34), pp. 139-150.
 O, O.B.O.D.E.H & C I, A.J.U.W.A. 2009. Evaluation of
Artificial Neural Network Performance in Predicting Diesel
Engine NOx Emissions. Applied Sciences, Engineering and
Technology. 33(4), pp. 642-653.
 Aydogan, H.A.S.A.N. 2015. Prediction of diesel engine
performance, emissions and cylinder pressure obtained using
Bioethanol-biodiesel diesel fuel blends through an artificial
neural network. Energy in Southern Africa. 26(2), pp. 74-83.