Scribd
Upload
70 views27 pages

Unconventional Manufacturing Techniques

The document discusses modeling material removal rate (MRR) and tool wear rate (TWR) in electrical discharge machining (EDM) using an artificial neural network (ANN). The objective is to program and train an ANN to predict MRR and TWR for EDM of AISI D2 steel. The ANN is trained on sample data and tested on new data, finding acceptable error between predicted and actual values. Further improving the ANN model and applying it to optimize EDM processes for reduced cost and waste is discussed.

Uploaded by

Sathya Narayanan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
70 views27 pages

Unconventional Manufacturing Techniques

The document discusses modeling material removal rate (MRR) and tool wear rate (TWR) in electrical discharge machining (EDM) using an artificial neural network (ANN). The objective is to program and train an ANN to predict MRR and TWR for EDM of AISI D2 steel. The ANN is trained on sample data and tested on new data, finding acceptable error between predicted and actual values. Further improving the ANN model and applying it to optimize EDM processes for reduced cost and waste is discussed.

Uploaded by

Sathya Narayanan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 27

ME4323 UNCONVENTIONAL

MANUFACTURING TECHNIQUES

NEURAL NETWORK MODELLING OF


EDM PROCESS

KRATIKA (ME16B114)
NIKHIL R (ME16B120)
SAHYA NARAYANAN (ME16B126)
SHRI VIGNESH (ME16B159)
MEENAKSHI SUNDARESWAR(ME16B149)
OBJECTIVE

• To program and train an ANN to predict the material


removal rate and tool wear rate of edm process for AISI
D2 Steel using a sample data and comparing the
estimated value with real time data and finding the error.
INTRODUCTION

• EDM is a manufacturing process where a desired shape


is obtained by using electrical discharges (sparks).
• Material is removed from the work piece by a series of
rapidly recurring current discharges between two
electrodes, separated by a dielectric liquid and subject to
an electric voltage.
• One of the electrodes is called the tool-electrode, while
the other is called the workpiece-electrode.
• The process depends upon the tool and work piece not
making actual contact.
MRR IN EDM

• MRR in EDM is defined as


• MRR= (electrode area * depth of cut)
time of cut
• But MRR depends on various parameters such as
• Current waveform, voltage waveform, tool resistance,
pulse on time, pulse off time, spark gap, dielectric
medium.
TOOL WEAR
ARTICIAL NEURAL NETWORKS

• ANN's are computing systems which are vaguely inspired


by biological neurala networks.
• It is not a algorithm,but rather a framework for many
different machine learning algorithms to work together
and process complex data inputs.
• An ANN is based on a collection of connected units or
nodes called artificial neurons, which loosely model the
neurons in a biological brain.
• The output of each artificial neuron is computed by some
non-linear function of the sum of its inputs.
ANN PERFORMANCE ANALYSIS

• Neural Architecture- The particular ways in which the the


given number of neurons are arranged.Here it is choosen
as MLP.

• Learning Algorithm- The training algorithm used to train


the data set is LMA.

• Number of Neurons- The neurons number in hidden


layer is varied as 6, 8,10 and 16. The weights of neurons
are varied in the process of testing.
THE 3 PHASES OF ANN
• Training data set- A training dataset is a dataset of examples
used for learning, that is to fit the parameters (e.g., weights). 70%
data points are used for this

• Validation data set- A validation dataset is a dataset of examples


used to tune the hyperparameters( i.e the no of hidden units).
15% data points are used for this

• Test data set- A test dataset is a dataset that is independent of


the training dataset, but that follows the same probability
distribution as the training dataset. 15% data points are used for
this
ANN OUTPUT PARAMETERS

• Mean Squared Error- It is defined as

• For multiple outputs, it is expressed as


• dnk- the desired output for exemplar n at neuron k of
output layer
• ank- the network output for exemplar n at neuron k of
output layer (i.e. the predicted output of neural network)
• K is the number of neurons in the output layer
• N is the number of exemplars in the data.
ANN OUTPUT PARAMETERS

• CORRELATION FACTOR- It is used to determine how well


the network output fits the desired output.
• It can be mathematically defined as

• For the neural networks the correlation coefficient is


defined as
• n- exemplar or run number

• an and dn - the network output and desire output


respectively at a particular exemplar, and are the data
mean of network output and desire output respectively
FINAL RESULTS

No of neurons r mse
6 0.990628 4.0914
8 0.997973 0.978745
10 0.996883 1.3272
16 0.99724 2.01249
MSE VS NO OF NEURONS
REGRESSION PLOT
ANALYSIS BETWEEN ACTUAL AND ANN VALUES
ANN PREDICTED AND ACTUAL MRR VS DATA
POINTS
ANN PREDICTED AND ACTUAL MRR VS DATA
POINTS
INTERPRETATION

Modelling and optimization of EDM process is a highly demanding research area in the
present scenario. Many attempts have been made to model performance parameters of
EDM process using ANN. For modelling generally ANN architectures, learning/training
algorithms and nos. of hidden neurons are varied to achieve minimum error. Hence
optimum number of neurons and the algorithm for achieving this has been performed. The
trained network is used in identification of MRR and TWR (Tool Wear Rate) for the test
data which was not utilised in training of the network and the predicted values are compared
to the experimentally observed values. The results have been discussed in this report and
further scope of extension of this project is also discussed at the end.
Conclusion and Future Scope
Conclusion:
(i)The ANN was created and trained for the sample set.
(ii)It was then run and compared for a new sample set.
(iii)The error was found to be in acceptable range and hence the program can be used for real time
purposes.

Further Scope:
(1) Since this project only considered MLP algorithm with varying neurons, we can use some other
algorithms like genetic algorithm and perform a similar analysis.
(2) This can be used for optimisation of the EDM process reducing cost and material waste.
(3) Surface roughness can also be added as output to this network and a relation can be developed.
Since we had only limited data source it has been restricted to MRR and TWR.
(4) The most important proceedings is we can do the inverse problem of finding the optimal
parameters for given MRR and Surface roughness values. If sufficient data points are
available/collected the network can be used for finding the cause for effect without any prior
knowledge about the inverse calculations.
THANK YOU !!!!

You might also like