Prediction of weldment mechanical properties in GMAW with robot-assisted using fuzzy logic systems

Welding operation decides the quality of product standards in all metal work products like automobiles, aerospace vehicles, and many more. The quality of the welding process is more reliable by automating the process with robots. In this research work, the GMAW operation is automated with the ‘Fanuc Robot Arc mate 100iC/12’ robot. The material characteristics such as ultimate tensile strength, hardness, and impact strength of weldments are predicted using a fuzzy system using triangular membership function (TrMF) and trapezoidal membership function (TMF). The simulated results are validated by comparing with experimental work, the experiments are designed using orthogonal array L18, and material characteristics are studied using fractography test. The fuzzy system is trained with experimental results using the IF-Then rule base with the help of the L18 orthogonal array. The inference system has predicted the accuracy rate of weldment mechanical properties, showing a lower error rate.


Introduction
Products like gates, fences, small kitchen appliances, vehicles, and even space travel vehicles utilise SS316 as raw material because of its higher mechanical strength, corrosion resistance, and shining appearance.SS316, Chromium has self-healing corrosion resistance, and Manganese provides maximum yield strength to the stainless steel.Welding is essential in all manufacturing industries: construction, aerospace, automotive, and nuclear.Gas Metal Arc Welding (GMAW) has found its application in welding pressure vessels and aerospace components [1].Above 70% of GMAW are performed by skilled welders.Both statements above convey that the industry is keen on implementing the six sigma process, and it is essential to automate the GMAW welding process.Also, according to the American welding society, by 2025, there will be a shortage of welding professionals around the globe.Welding robots are the prime source of compensation for the skilled professionals of welding.
Wang et al [2] performed a numerical investigation on the process parameters in GMAW by performing 3D model interactions among the process parameters such as arc plasma, droplet and external magnetic field.Huang et al [3] performed quality diagnosis in GMAW by varying the arc signal using empirical mode decomposition and extreme learning machine; finally, the combination has shown reliable results in identifying weld quality.Thompson et al [4] performed image analysis in GMAW using deep learning and machine learning techniques for predicting the bead geometry and verified with experimental results.Kamble et al [5] studied GMAW process parameters by formulating a mathematical model for estimating the values and comparing them with experimental results.The process parameters of GMAW were predicted through thermomechanical simulation using ANSYS software on AISI 430 steel and compared with experimental results.
Oliveria et al [6] proposed a detailed forecast of the GMAW and cold metal transfer welding process based on lifecycle analysis that precisely defined the welding characteristics.Yong et al [7] performed a stability analysis of GMAW using a genetically optimised support vector machine.The study has given the stability of shortcircuiting is closely connected with entropy distribution.Dongsheng et al [8] investigated the heat transfer in hybrid KPAW-GMAW process in three-dimensional simulation and concluded that proposed Marangoni force and arc shear stress are dominant forces.
The structure of the manuscript is as follows.Section 2 illustrates the related works carried using fuzzy logic systems.Section 3 discussed the problem definition of the work.The experimental setup is described in section 4. The architecture of fuzzy system formulation is narrated in section 5. Section 6 explains the result and discussion of the fuzzy system prediction.Finally, the conclusion section explains the outcomes of research followed by references.
This research attempts to investigate an experimental robot-assisted gas metal arc welding (GMAW) welding process on SS316 material.

Related works in fuzzy logic systems
Latha et al [9] have predicted the surface roughness of the drilling hole using fuzzy logic systems, which has high intelligent reasoning capabilities.Aamir et al [10] optimised the drilling parameters and processes using the fuzzy and Taguchi method.Ramesh et al [11] have predicted the parameters of turning operation such as flank wear, surface roughness using fuzzy rule-based modelling.Fuzzy logic is applied in selecting the cutting conditions during turning operations [12].The performance characteristics of drilling operations like thrust force, torque are predicted using fuzzy logic systems [13].The useful life of the end mill cutter is predicted [14] using fuzzy logic techniques, with remarkable tool life predictions.The influential phenomena [15,16]for drilling operation are predicted using the fuzzy logic system by modelling vibration-assisted operation's kinematic.
Khamari et al [17] optimised the welding parameters of GMAW and SMAW with mild steel for characterising the mechanical and microstructural properties.Amruta et al [18] developed a new approach by combining fuzzy-regression and enhanced teaching-learning optimisation process, which has an optimal setting of welding parameters and validated experimentally by performing welding operation with MOTOMAN MA 1440.Karuthapani et al [19] studied the relationship between GMAW parameters and macrostructural characteristics and developed a fuzzy logic model to predict the effect of the flat electrode.Li et al [20] proposed a novel controlling mechanism for improving the fabrication accuracy rate with fuzzy logic inference.Xie et al [21] developed a comprehensive evaluation of welding quality using the fuzzy logic controller.The readers can find further detail on the arc-based welding process [22].

Problem definition
Manual Controlling of the GMAW process parameter was a tedious task as the uncertainty had a greater impact on the welded joint's quality.Also, predicting GMA's performance is always a challenging job, which helps to plan efficient welding operations and automate the welding operation with robot assistance.In this research, an attempt was made to weld SS316 using Fanuc Robot Arc mate 100iC/12 by varying gas flow rate (l min −1 ), wire feed rate (m min −1 ), travel speed (cm min −1 ) and current (A).The obtained experimental were trained using fuzzy logic systems to predict the performance of GMAW.The predicted results were compared with the attained experimental value to validate the model.

Experimental setup
GMAW operation is performed with specimen material of SS316, and the size of the specimen is 120 mm × 60 mm × 3 mm.Both the filler material and base metal are in the same material SS316, and their chemical composition are in table 1. Examining the GMAW operation are carried by varying the input parameters such as gas flow rate, current, travel speed, and wire feed rate (table 3).All four parameters have three variation levels.The variation in each parameter will affect the output parameters such as ultimate tensile strength (UTS), hardness and impact strength.The experiment study is performed by formulating the experiment sequence in the L18 orthogonal array, and it is enumerated in table 4.

Prediction of GMAW parameters using fuzzy logic
The output parameters of GMAW are predicted using fuzzy logic systems, and the fuzzy controller is trained with the rule base, as shown in table 5.The rule base has trained the fuzzy system to predict the output    .Centroid type fuzzification is adopted to perform inference in the fuzzy logic system.Takagi-Sugeno-Kang model type is adopted in fuzzy inference system.The mathematical formulation of both membership functions is shown in table 6.The coordinate points of both TrMF and TMF are shown in tables 7 and 8.

Rule base
The fuzzy inference system is trained with a rule base, and the rule base is designed with IF-Then condition based on this rule-based will train the inference system the predicts the output parameters.The rule base represents the performed experiments results as per L18 orthogonal.Hence the rule base is defined with linguistic variables, and the details of each parameter are listed in tables 6 and 7.

Results and discussion
The simulation is performed with Python 3.7, Spyder 4.2.5, and 'skfuzzy' toolbox is to simulate fuzzy inference systems.The workstation configurations are Windows 10 OS, and 16 GB of RAM

UTS
The specimen's ultimate tensile strength is predicted using a fuzzy system with two different membership functions such as TrMF and TMF are shown in figure 6.In TrMF, higher error rates are found in 1.1%, 1.2%, 1.3%, and 1.1% in the second, seventh, fourteenth, and fifteenth experiments.The other experiments have lesser error rates.In TMF, maximum error experiments are second has 1.1%, seventh has 1.2%, fourteen has 1.3%, and fifteen has 1.1%.Both memberships have a similar pattern of prediction.While welding the material alpha and delta ferrite is formed, delta ferrite is at a higher temperature and remains the same during the cooling phase.This is phase transformation leads to avoiding hot cracking and solidification cracking during the welding [23].

Hardness
The simulation results for predicting hardness parameters with TrMF and TMF are shown in figure 7.In TMF highest error found in 1st, 4th, 7th, 8th, and 9thare 2.3%, 1.3%, 1.5%, 1.6%, and 2.1% respectively.Similarly, TrMF is found in both error rate and experiment numbers.Phase transformation occurs in the welded zone from alpha ferrite to martensite [24].In the solidification process, normally, stainless-steel materials losses their original strength in the weld zone because of the grain structure.But in the SS316 material, due to the carbon content and cooling, martensite is formed in the fused zone, increasing the material's hardness.

Impact strength
Impact strength of the specimen is also predicted with the fuzzy system, and the prediction has shown in figure 8, similarity in both the membership functions, and the highest error rates are also similar, and the error rates are 2.2% in 4th and 6th, 1.5% in 17th and 18th experiments.In contrast, the other experiments have trivial error rates.The coarse grain structure formed in the HAZ is due to alpha ferrite formation.However, refined grains are included in the weld zone due to the formation of martensite and austenite, which increases the material's toughness [25].

Fractographytest on the specimen
The fractography of the tensile specimens is observed to analyse the fracture behaviour of the SS316 weldments.
The fractography of specimen welded using a gas flow rate of 10 l min −1 , current of 120 A, travelling speed of ; , , , 45 mm min −1 , and wire-speed of 5 mm min −1 is shown in figures 9(a) and (b).This test specimen appears more brittle due to the low welding current.The brittleness increases due to voids in the fractured region, so the failure occurs during the tensile test.This happened due to the low melting in the weld zone because of the weld current and wire feed rate deficiency.In general, brittle fracture mode propagates along the trans-granular grains in metals.The brittleness increases due to melting pools formed in the fused area, and also, the solidification temperature range is minimum in the centre line of the weld.These causes because of the cooling from 1400 °C to 800 °C, the shrinkage takes place in the HAZ, and WZ sets the tensile residual stress in the material.Figures 9(c) and (d) show the fracture surface of specimen welded using a gas flow rate of 15 l min −1 , the current level of 140 A, travelling speed of 35 mm min −1 and wire speed of 4 mm min −1 .Dimples, shear lip formation and plastic flow are the observations made in the fracture surface, and this offers better elongation of metals during the tensile test.High current and reasonable travelling speed provide an expected level of metal deposition in the weld regions and favour high tensile strength.The increased flow of current in the weld zone leads to the material's ductility and reduces the solidification cracking.This happened because ferrite content formed correctly in the fused zone.

Comparison of results
The simulation results of both membership functions are compared in terms of average error in prediction for all the experiments performed.Both membership functions have shown similar error rate predictions in UTS     11

Conclusion
The attempt to predict the performance of GMAW using the fuzzy system with TMF and TrMF has shown a lesser average error rate in all three parameters, such as UTS has 0.56%, hardness has 0.83, and impact strength has 1.08%.The error rate in the predicted results is calculated by comparing with the experimental results.Both TrMF and TMF has shown similar error rate in all three output parameters, whereas TrMF has gained the advantage of having the lesser parameter to represent.Hence, the computation cost is lesser than TMF.The fuzzy space of all three output parameters is classified with nine linguistic variables, which increased the fuzzy learning rate to the inference system.The rule base is formulated based on the L18 orthogonal array to train the fuzzy inference system more accurately, which has reduced the error rate in all three parameters.The error rate of all 18 experiments for each parameter is compared, with the highest error rate as 1.3% in 14th, 2.3% in 1st and 2.2% in 4th experiments of UTS, hardness and impact strength, respectively.
Future research can be carried with the error rate in the parameters being reduced by implementing hybrid intelligence systems by combining fuzzy logic and machine learning techniques.Machining parameters can be predicted in other manufacturing operations like milling and grinding operations.

Table 6 .
Parameters of membership function.

Table 1 .
Chemical composition of SS316 material.
4.1.Design of experiments GMAW operations are performed with the 'Fronius TPS 400i MIG' welding machine assisting with robot 'Fanuc Robot Arc mate 100iC/12'.The technical specifications are given in table 2, and the setup is shown in figure 1.

Table 3 .
Input parameters of GMAW and their variation levels.
parameters of GMAW.The rule-based is designed as per the results of the experiments of the L18 orthogonal array.The architecture of the fuzzy system is shown in figure 2. Centroid fuzzification is carried to fuzzify the parameters of GMAW.The specimen of experiments carried as per L18 orthogonal are demonstrated in figure 3.

Table 4 .
Design of experiments of GMAW and its results.

Table 5 .
Rule base for fuzzy controller.

Table 7 .
b, c, &d-Coordinate points of trapezoidal membership function Coordinate points of TrMF.

Table 8 .
Coordinate points of TMF.
Mater.Res.Express 8 (2021) 126524 P Devendran and P Ashoka Varthanan and hardness, whereas impact strength has slight variations.A detailed error rate comparison is itemised in table 9.