**Stage 2: Alloying of AISI P20 mold steel with the use of powder metallurgy electrodes of titanium and aluminium has been carried out in a hydrocarbon oil dielectric medium**

**B. Mesh sensitivity analysis**

**7.6 Assessment of the developed integrated FEM-ANN model**

architecture are shown in Figure 7.16. It can be noted that in all the cases, the R-value is
above 0.85, which is acceptable. Therefore, the 3-10-1 network architecture was
considered as the optimal network configuration for accurate prediction of *F**A*. The
performance of this network was verified by using a set of processing conditions that were
not used in the training.

**Figure 7.16 **Regression plots for training, validation, testing, and all the dataset for 3-
10-1 network architecture** **

network were employed. Computed alloyed layer thickness values and corresponding
process conditions are listed in Table 7.11. Initially, by using the optimal network, the *F**A*

values have been computed for the chosen 8 datasets. Then, by employing these *F**A* values
and the respective processing conditions as input to the developed FEM model, the
corresponding values of alloyed layer thickness were computed. The computed alloyed
layer thickness using the FEM-ANN model was then compared with the experimentally
determined alloyed layer thickness for similar process conditions. The absolute %
deviations were computed and tabulated (Table 7.11). It was observed that the average
percentage deviation was 6.55 %, and the range of the average deviation ranges from 1

% to 14 %, which is considered to be acceptable. Thus it was noted that the developed FEM-ANN could be adopted to accurately and quickly compute the alloyed layer thickness. A similar methodology can further be employed for any tool-work material combination. Therefore, the developed FEM-ANN integrated model was found to be effective and robust as the developed model could successfully compute the expected alloyed layer thickness for the considered process parameters. Further, the present study added more insight in the mechanism of electric discharge alloying process.

**Table 7.11 **Assessment results using 3-10-1 network
Sl.

No.

Dielectric medium

Pulse on-time (µs)

Discharge current (A)

ANN
predicted
*F**A*

Alloyed layer thickness (µm)

% Deviation Numeri-

cal

Experimen- tal

1 1 706 6 0.185 37.89 38.54 1.68

2 1 706 8 0.178 42.94 41.55 3.34

3 1 856 12 0.198 59.13 51.81 14.13

4 1 1006 8 0.165 37.88 34.98 8.29

5 2 706 10 0.185 50.25 46.42 8.25

6 2 856 10 0.172 47.05 45.73 2.88

7 2 1006 12 0.158 46.35 53.11 12.73

8 3 546 8 0.142 32.90 33.28 1.14

Average deviation 6.55 %
**7.7 Summary **

In the present work, an integrated FEM-ANN model has been successfully developed to accurately predict the alloyed layer thickness in electric discharge alloying of AISI P20

mold steel using powder metallurgy electrode of titanium and aluminium at different
processing conditions such as varying discharge current, pulse on-time, and dielectric
medium. The total time required to predict the fraction of energy and the alloyed layer
thickness is approximately 5 mins. Very scant work is reported on the prediction of
alloyed layer thickness using artificial neural networks and hard computing methods
together. Literature reports prediction of material transfer rate and average alloyed layer
thickness in the EDA process using ANN. However, it needs large input data as the
developed model was solely dependent on the empirical relation resulting in high
experimental cost. In the present work, the developed model has achieved the accurate
computation of the alloyed layer thickness with fewer data. This has been possible by
incorporating the optimal fraction of energy distribution value in the developed FEM
model. The fraction of energy distributed to the workpiece is one of the important process
parameters in modeling the electric discharge phenomenon. Works have been reported in
the inverse computation of energy distribution factor *F**A *by determining the surface
temperature or the crater radii. However, in the present work, *F**A* was computed from
experimentally determined alloyed layer thickness. The alloyed layer thickness was
computed by considering accurate values of fraction of energy distribution to the
workpiece, *F**A*. These values were computed by using the inverse estimation method and
the ANN-based model. Following important conclusions were drawn from the study.

The neural network of 3-10-1 architecture was found to be the optimum network.** **

The developed methodology suggests that the fraction of energy *F**A* varies from 0.129
to 0.215. This can be employed in the thermal analysis of the electric discharge-based
manufacturing processes. ** **

The performance of the developed FEM-ANN was verified by carrying out the
experiments. It was found acceptable with an average prediction deviation of 6.55 %.** **

The present work facilitates a simple and quick methodology for accurate prediction of the alloyed layer thickness for complex manufacturing processes such as EDA.

This provides an efficient and economical alternative to the costly, tedious, and time-
consuming experimental work**. **

CHAPTER 8

**Conclusions and Future Scope **

The main objective of the present work was to enhance the surface characteristics of AISI P20 mold steel in terms of its hardness, corrosion, and wear resistance by electrical Discharge Alloying. In view of this, the work was planned to investigate the alloying of titanium, aluminium, and nitrogen over AISI P20 mold steel. For this, a powder metallurgy tool of titanium and aluminium was used, and three different dielectric media viz. hydrocarbon oil, deionized water, and urea mixed deionized water were used. Further, to study the underlying process mechanism of EDA, a physics-based numerical model has been developed to predict the alloyed layer thickness. The alloyed layer thickness was computed by considering accurate values of the fraction of energy distributed to the workpiece. This fraction of energy distribution was computed by the inverse estimation method. Thereafter, the developed model has been integrated with ANN to develop a robust predictive model. The overall research work has been carried out in the following major stages.

Experimental investigations into electrical discharge alloying of titanium and aluminium with AISI P20 mold steel in the presence of hydrocarbon oil dielectric medium

Experimental investigations into electric discharge alloying of Ti and Al on P20 mold steel with a water-based dielectric medium

Characterization and assessment of the alloyed workpieces in terms of its wear and corrosion resistance behavior.

Computation of alloyed layer thickness in electric discharge alloying by inverse estimation of energy distribution.