The Use of Fuzzy Logic in Predicting Percentage (%) Dilution of Weld during Tig Welding Process

Welding operation of metals, gives rise to high temperature that results in melting of mating parts. The final composition of the joints formed in terms of its microstructure and properties at the fusion zone depends greatly on the degree of dilution of the weld. With an expert prediction technique, it may be possible to predict even before weld, the integrity of weld joint from the proposed process parameter. The aim of this study is to predict the percentage dilution (%D) of TIG mild steel welds using fuzzy logic. In this study, the weld specimen was produced using the TIG welding process guided by the central composite experimental design and thereafter percentage dilution (%D) was measured and fed to the fuzzy logic software. The process parameters include the voltage, current, gas flow rate and welding speed. The results obtained showed that the fuzzy logic tool is a good predictive tool and the model developed has proven to be very efficient in handling works of this nature, thereby saving time, energy and money wasted in pre-welding procedures. It would be encouraging to compare other quality parameters with process parameters to see how it can further help in quality improvement.


Introduction
predicted bead geometry for lap joint in gas metal arc welding (GMAW) process. They conducted experiments based on five process parameters to obtain bead geometry using GMAW process. From the study, it was inferred that welding voltage, arc current, welding speed and welding angle have large signif- icant effects on bead geometry. [2] optimized parameter for Dissimilar Pipe Joints using TIG welding process. Good weldability of metals can be linked to the quality of their bead geometry and welded joint geometry. [3] investigated effect of Process Parameters on Depth of Penetration in Gas Tungsten Arc Welded (GTAW) 202 Grades Stainless Steel Plates. [4] optimized parameter for weld strength of metal inert gas welding and tungsten inert gas welding. [5] investigated the effect of welding speed on the tensile strength of the welded joint.
While different efforts have been put to improve the weld qualities, it must be noted that the most economical ones are the theoretical or computer aided simulations like in the case of [3] that used response surface methodology (RSM) to establish quadratic relations between the welding process parameters and bead geometry, for depositing 316 L stainless steel onto structural steel, using automated submerged arc welding (SAW) and MIG welding, separately. [6] studied on the use of the grey-fuzzy logic based on orthogonal array for optimizing the electrical discharge machining process with multi response characteristics. [7] proposed an intelligent technique, adaptive neuro fuzzy inference system (ANFIS) to predict the weld bead width in the submerged arc welding process for a given set of welding parameters. They designed the experiments using Taguchi's principles and used the results to develop a multiple regression model. Data obtained from multi regression model were used to train the intelligent network. They found that the proposed ANFIS is flexible, accurate in predicting bead width. The quest for quality improvement of welds cannot be over emphasized as we often encounter failure of structural materials at the points of their joining which has posed a serious challenge to the fabrication industries. However, the quality of these welds is determined by the process parameters which significantly enhance the strength, reliability, and integrity of structural materials [8]. The mechanical metallurgical features of the weldment are subject to the weld bead geometry, which is directly interconnected with welding process parameters [9]. To have a better control over weld quality and bead geometry, it is necessary to establish the relationships between welding parameters and weld bead dimensions. In this light, if software is employed, it may be possible to predict the response to some degree of accuracy and vary the process paramenter to achieve a desired response [10].

Materials
The Tungsten Inert Gas (TIG) machine was used, thirty experimental runs were carried out, each experimental run comprising the current, voltage, welding speed and gas flow rate, used to join two pieces of mild steel plates measuring 60 mm × 40 mm × 10 mm. The percentage (%) dilution was measured respectively.
The welding process parameters are shown in Table 1.

Modeling and Prediction of Dilution Using Fuzzy Logic
The general structure of fuzzy inference systems is shown in Figure 1. Fuzzy  In this study we developed a fuzzy logic system to predict the percent dilution based on four input variables namely: voltage, current, welding speed and gas flow rate.

Defining the Linguistic Variables and Terms
Linguistic variables are the input or output variables of the system whose values The range of the input and output variable were extracted from the welding process parameters limits experimental design summary presented in Table 1. The fuzzy logic tool box that defines the input and output variables is presented in Figure 2.

Converting the Crisp Variables into Fuzzy Sets
To convert the crisp variables (actual experimental data) into fuzzy sets, adaptive neuro fuzzy inference system (Anfis) was employed to generate a fuzzy inference system (FIS). To generate the FIS structure, the raw data were presented as input parameters in anfisedit tool box. A section of the raw data presented to MATLAB as input parameters is presented in Figure 3. Figure 3 shows the crisp (raw) data for voltage, current, welding speed, gas flow rate and dilution (%). Since fuzzy do not accept the crisp data, the crisp data were then sent to adaptive neuro fuzzy (anfis) for possible conversion into fuzzy sets. The process of converting the crisp data into fuzzy sets is called fuzzification. For this problem, the fuzzification step was done using anfis as presented in Figure 4. Figure 4 shows the interface of adaptive neuro fuzzy (anfis) containing the crisp data for voltage, current, welding speed, gas flow rate and dilution (%) in    owing to its simplicity as presented in Figure 5. Figure 5 shows the number of membership function that was assigned to each linguistic variable and the type of membership function that was selected for each input and output variable. As observed in Figure 5, five membership functions were defined for each input and output variable while the triangular and the constant membership function were selected. Summary results of membership function and membership sets for voltage, current, weld speed, gas flow rate and percentage dilution are presented in Table 2.

Create Membership Function (Initialization)
For the critical rules constructed for predicting dilution using fuzzy logic, the simplified form of Figure 14 is presented as follows:         1) If voltage is low and current is high and welding speed is high and gas flow rate is high, dilution is very low.
2) If voltage is moderate and current is moderate and welding speed is moderate and gas flow rate is very high, dilution is low.
3) If voltage is high and current is high and welding speed is low and gas flow rate is low, dilution is moderate.

4) If voltage is high and current is low and welding speed is high and gas flow
rate is high, dilution is high.

5) If voltage is very high and current is moderate and welding speed is mod-
erate and gas flow rate is moderate, dilution is very high.

6) If voltage is low and current is high and welding speed is low and gas flow
rate is low, dilution is moderate.

7) If voltage is low and current is low and welding speed is low and gas flow
rate is low, dilution is low.

8) If voltage is low and current is low and welding speed is high and gas flow
rate is high, dilution is low.

Construct the Rule Base (Initialization)
In a fuzzy logic system, a rule base is constructed to control the output variable.
A fuzzy rule is a simple IF-THEN rule with a condition and a conclusion. Based  Table 2, eight critical rules were constructed to predict the dilution based on fuzzy logic. Figure 14 shows the fuzzy rule editor containing the eight critical rules constructed for this problem. Figures 15-22 show the predictions that were made using fuzzy logic systems.

Predicting Dilution Using Fuzzy Logic
From the result of Figure 15, it was observed that; for a voltage of 20 volt, current of 160 Amp, welding speed 170 mm/min and gas flow rate of 14 L/min, the predicted dilution was 43.10 mm.
From the result of Figure 16, it was observed that; for a voltage of 22 volt, Figure 14. The critical rules constructed for predicting Dilution using fuzzy logic. From the result of Figure 17, it was observed that; for a voltage of 24 volt, current of 160 Amp, welding speed 150 mm/min and gas flow rate of 12 L/min, the predicted dilution was 70.70 mm.
From the result of Figure 18, it was observed that; for a voltage of 24 volt, Figure 16. Prediction of Dilution using fuzzy logic. From the result of Figure 19, it was observed that; for a voltage of 26 volt, current of 150 Amp, welding speed 160 mm/min and gas flow rate of 13 L/min, the predicted dilution was 98.20 mm.

Discussion of Results
The surface plot which shows the relationship between the input and the output variable is presented in Figure 23 & Figure 24.
Result of Figure 23 and Figure 24 shows the dependence of the input variables on the output variable (dilution (%)). It shows clearly that any change in the input variable will result in a significant change in the output variable.
The randomized selected performance evaluation of fuzzy logic and experimental means in predicting dilution are shown in Table 3.
Learning parameters on the accuracy of weld predictions were studied along the variations of the input parameters for fuzzy logic and it was able to produce a model capable of predicting percentage weld dilution beyond the range of the given parameters. The randomized selected results obtained from fuzzy logic for minimized percentage weld dilution are presented in Table 3 in relation to the actual experiment which indicates that result of Fuzzy logic were very close to results of the actual or observed experiment. Table 3 also shows that it is possible to predict the percentage weld dilution with only the process parameter, we see that for a randomized serial number (R/N) of 7 in Table 3    This research work has thrived in developing an optimization and prediction of welds of extremely high quality of TIG welding process using fuzzy logic through which the effects of their various process parameters and their interactions were determined and predictions made on expected quality of the weld at known process parameters.

Conclusion
A novel concept of an intelligent model has been developed to predict welding process parameters (current, voltage, welding speed and gas flow rate) and bead parameters (%D) for improved quality welds using fuzzy logic. The results of this study will help reduce the cost of expensive analytical methods employed during welding operation and it will help fabrication industries to maximize the quality of their products with minimal stress and save them time normally applied to do a trial and error work pre to welding.