Machining Characteristics Investigations of DSS-2205 Using RSM – ANN and Gray Relational Analysis

DSS has low machinability characteristics due to its high strength, machining is complicated, and careful attention is required when selecting machining parameters. The main criteria discussed in this paper concern the turning optimization parameters and machining time reduction of DSS 2205 as the work material. The input parameters are cutting velocity, feeds, cutting depth, and tooltip nose radius of the cutting tool. The design of experiments methodology is employed to design the experiments using Design-Expert V12 software. The second-order mathematical model was developed, and analysis of variance was performed to analyze the performance characteristics to recognize the critical variables in ﬂ uencing the output parameter. An arti ﬁ cial neural network (ANN) backpropagation algorithm using MATLAB software was used to develop the mathematical model and optimize the output. The model was developed, and the results were optimized using MATLAB software ’ s ANN back propagation method to ﬁ nd the best possible solutions. The generated models were signi ﬁ cant based on the analysis of variance and the R -squared value, and these results indicate that the cutting velocity is the most critical factor. For a low machining time, the cutting velocity should be between 100 and 140 m/min, and the tooltip nose radius should be 2.8mm. The optimal parameter settings are validated by performing a lower is better con ﬁ rmation test using gray relational analysis (GRA). The GRA exposed the lower machining time at a cutting velocity of 140 m/min, rate of feed of 0.5mm/rev, cutting depth of 0.5mm, and tooltip nose radius of 2.4mm. The predicted values were close to the experimental values, and the result indicates the optimal level of the highest GRA grade of the machining variable.


Introduction
Germany and Sweden invented the 22% Cr composition of duplex grade S32205 in the 1970s.It is essential to maintain nitrogen concentration; the heavily alloyed duplex grade was designed in the 1980s as the so-called super duplex grade.After the advancement, the scientists developed a lean alloy grade, DSS 832304.Alloying with nitrogen inspired the scientists to introduce many DSS grades.Despite its high yield strengths and high ductility properties, the producers and end-users have more significant restrictions during machining and welding [1].DSS has the best mechanical properties and the highest resistance to chloride corrosion due to its progressive ductile-brittle transition and combined austenite and ferrite microstructure [2].Palanisamy and Selvaraj [3] used cast DSS to assess the roughness of the workpiece and suggested that changing the grain orientation of the working materials had the most significant effect on the surface.Krolczyk et al. [4] examined the cutting performance of cutting tools MM 2025 and CTC 1135 during the machining of DSS 1.4462 material under dry and wet circumstances.The result revealed a minor difference in the effect of cooling on the material surface of the tools.The focus of the research is to forecast the roughness of the DSS material during the turning operation.The author derived the equations, and the results emphasize that the feed rate was the most critical factor affecting roughness [5].Krolczyk et al. [6] derived a technique to estimate the life of the cutting tool during the turning of DSS material.The response surface methodology (RSM) technique creates a second-order model to predict equations for tool life and the results; cutting velocity was the crucial factor affecting the tool's life.The facing operation is performed under a constant cutting speed using austenitic, standard DSS, and super DSS material.Optimization was carried out using the Taguchi L16 orthogonal array technique.Koyee et al. [7] explained that the snarl and ribbon chips were created at lower feeds and higher cutting depths while milling super DSS and standard DSS forms.Fragments produced by milling austenitic stainless steel are machine-friendly, reducing ribbon, knotted, and flat helical chips.
Using the infinite focus measurement machine, Krolczyk and Legutko [8] investigated the surface texture and tool wear on DSS material.The author explained that when the feed rate increases, surface roughness also increases, and dry cutting enhances roughness and material ratio characteristics.The STD DSS, SDSS material, and carbide inserts coated with CNMG120408-QM 2025 were used as the cutting tool for experimentation.The author used a multiobjective bat algorithm (MOBA) to optimize the turning topographies.The MOBA regularly predicts a set of optimal solutions [9].Analysis of variance (ANOVA) was used to investigate the performance characteristics of three DSS materials utilizing the mathematical model.Multiattribute decision-making has been used, including graph theory and matrix approach and analytical hierarchy process-technique for order preference by similarity to ideal solution, for optimization purposes [10].Koyee et al. [11] developed prediction model experiments in dry and cooling lubricating conditions and compared them with cutting fluids and dried-out turning to lengthen tool life.
To determine the most effective machining method and cutting circumstances, Krolczyk et al. [12] employed DSS 1.4462 using the optimization techniques of ANOVA and Taguchi's orthogonal array L18.The influence of cutting speed, feed rate, depth of cut, and tool nose radii are studied using Taguchi's approach and ANOVA method to minimize surface waviness and material removal rate (MRR) [13].The effects of surface and tool wear during the turning of DSS material in a traditional turning machine were investigated by Liew et al. [14].Surface roughness decreased due to chilled air coolant instead of predictable coolant as the chilled air coolant's temperature fell, consequently affecting the surface roughness.Nomani et al. [15] explored the machined surface process and machinability of different components, specifically DSS alloys, SAF 2205 and SAF 2507, using scanning electron microscopy and optical microscopic examination.Pawan and Misra [16] considered DSS material for experimentation to predict roughness.To assess the surface waviness of the DSS material, Selvaraj [17] used the Taguchi approach.This study focuses on the impact of physical vapor deposition coatings on surface roughness and residual stresses at the cutting temperature during the DSS 2205 turning.The study investigated the residual stresses and surface unevenness impacted by coated and uncoated carbide tools.Sonawane and Sargade [18] suggested that the surface roughness of a machined surface can be improved by increasing the cutting speed.To measure surface roughness and machining time, experiments are conducted on duplex 2205 using a CNC lathe and carbide tip tool material under hard turning.A second-order mathematical model was developed using RSM, and performance characters were evaluated using the ANOVA method.The MOGA method was employed to achieve the most practical responses [19].Kara et al. [20] experimented with AISI D2 hard turning under a cryogenic environment to assess the surface roughness and tool wear.
The reality of Duplex stainless steel is that it is resistant to chloride stress corrosion cracking.DSS material is twice more robust, low ductility than other austenitic grades.Duplex stainless steels are primarily utilized in specialized applications due to their limitations, which include poor formability and machinability and a more sophisticated metallurgical manufacturing process than ferritic, austenitic, and martensitic stainless steels.Therefore, careful consideration should be taken during the machining process.The literature indicates that the artificial neural network (ANN) heuristic method is suitable for enhancing the response optimization processes during the machining of Duplex grades.Even though numerous studies have been conducted, several types of research are available to predict surface roughness, tool wear, cutting force, and machinability analysis by using various optimization techniques during the turning of Duplex material.However, there is limited research available to optimize machining time using ANN on DSS material, and there is a significant gap in acknowledging and predicting machining time during the turning of DSS.Experimentation was conducted to overcome machining challenges and solve the critical concerns of industry specialists.The experiment was planned following the design of experiments (DoE) of central composite rotatable design (CCD).RSM and ANOVA were used to develop the mathematical models.The ANN, a heuristic optimization technique, was used.

Experimentation Details
The objective of this study was to determine the ideal machining conditions by estimating the optimal machining time during the turning of Duplex stainless steel (DSS-2205-ASTM A276) and by evaluating the impact of the cutting parameters.The cutting operation was carried out using a conventional Kirloskar Lathe using a coated carbide insert tool (Tungaloy-SNMG 120 408 MT AH925).The Duplex-2205 round bar of length 60 mm was used for experimentation.The parameter considered is cutting velocity (V c ), rate of feed (F z ), cutting Depth (D c ), and tooltip nose radius (R tip ).The experiments are designed per the central composite design, consisting of four-factor, five levels of the experiment using Design Expert Version 12 (stat case).The sample workpiece was machined to 20 mm, and a stopwatch was used to record the machining time.The Design for experimentation is shown in Table 1.The experimental results are tabulated in Table 2.

Identification of Design Matrix
The central composite experiment, i.e., CCD of RSM, is designed per the guidance given by Okouzi et al. [21].Each component is allocated to five equally spaced values, typically coded as per the procedure of Box and Draper [22].The experiment consists of four variables, five levels of CCD, 2 Mathematical Problems in Engineering and experimental sets of actual states are shown in Table 1.
The cutting condition ranges are selected from The IMOA Practical Guidelines Handbook [23].Trial runs using the specified parameters generated to precise projected values given in Tables 1, the upper bound (+2) and the lower bound (2) levels of each of the four independent variables.The interpolation method was used to determine all intermediate variables levels 1, 0, and +1.A four-factor CCD with 30 sets of coded conditions of full replication, where 24 are noncenter points and 6 are center points, was chosen as the design matrix to carry out the experiments using the DoE technique.
Table 2 shows the 30 sets of coded trail experiments.

Regression Equations
Karthik et al. [24] conducted the regression analysis based on the DoE considering spindle speed, feed rate, and depth of cut under dry and wet conditions.According to the set of techniques provided by the authors.The 2nd-order quadratic mathematical regression models were developed for the parameters as follows: Regression equations are used to predict the machining time.

Results and Discussion
The Design-Expert V12 software is a statistical tool used to conduct experiments based on the DoE and examine the impacts of the response variable.The design matrix used to carry out the experiments is shown in Table 2.The ANOVA was performed to verify the adequacy of the model.The findings of an ANOVA used to predict T m are shown in Table 3. Table 4 shows the percentage contribution of ANOVA.
The F value is 2.96, which is significant, and the P value is 2.26%, which may occur due to noise.The lack of fit (LF) F value is 0.2787, showing that the models' terms are not significant and have a noise of 95.54%.

Analysis of 3D Interaction Effect
The impact of process parameters cutting velocity (V c ), rate of feed (F z ), cutting depth (D c ), and Tooltip nose radius (R tip ) on machining time (T m ) is addressed below.
The interaction effect of cutting velocity and rate of feed on machining time is shown in Figure 1.The cutting velocity versus the rate of feed over machining time, as shown in Figure 1, has a significant influence on the machined surface and tends to increase machining time in hard turning.Figure 1 illustrates that when cutting velocity increases, machining time also increases; the machining time is minimum between 100 and 140 m/min.The rate of energy lost, plastic expansion, and friction increases as cutting velocity increases.When the feed rate increases, the machining time also increases proportionally, as illustrated in Figure 1; this is due to the larger cutting zone area between the tool and the workpiece, so a large amount of heat is developed at the tooltip, propagating the tool wear.The results are verified using the ANOVA table.The result concludes that the cutting velocity is the most influencing variable compared to other parameters.
The cutting depth plays a vital role throughout the machining, as shown in Figure 2. The cutting depth increases and the machining time increases; as a result, the quantity of workpiece materials is reduced.
Figure 3 shows the machining time relationship between cutting velocity and tooltip nose radius.The 3D plot in Figure 3 shows the machining time is high when the nose radius is at 2.8 mm and low at 0.8 mm. Figure 4 depicts the relationship between predicted and actual values.It helps to identify the observations by showing a graph of observed values vs. predicted values.The data points are evenly divided along the 45°line.

ANN Optimization
Hariche et al. [25] demonstrated that a mathematical model that attempts to simulate the structure and functions of biological neural networks is called an ANN.In recurrent ANN, information travels in one direction and the opposite direction in artificial outputs.An ANN has three layers: source, hiding, and destination.According to the author, the source and destination layers are called nodes, while the hiding layer links the origin and destination [26].Rajesh et al. [27] developed a predictive model using the ANN technique to optimize the machining parameter during the turning of Inconel 625.To reduce the number of experiments and timeconsuming trials, five learning techniques were utilized in  [29] used ANOVA and RSM to develop a prediction model to forecast surface roughness during the milling of AA6061 alloy.The ANN model was employed to identify the optimum machining conditions and evaluate the cutting parameters' impact.Manoj et al. [30] suggested that the ANN technique was the most effective method for the prediction of response parameters and also reduced the error percentage.ANNs may solve complicated real-world issues by processing fundamental building blocks in nonlinear, distributed, parallel, and local ways.An ANN model is developed using the MATLAB NN tool.

The use of ANN to reduce machining time is described:
To Minimize : within limits, (i) 90 m/min ≤ V c ≤ 120 m/min, (ii) 0.18 mm/rev ≤ F z ≤ 0.36 mm/rev, (iii) 0.20 mm ≤ D c ≤ 0.60 mm, (iv) 0.40 mm ≤ R tip ≤ 0.80 mm.Mathematical Problems in Engineering Subhash et al. [31] investigated the machinability and unstable process-induced chatter vibrations of dry-turning SDSS.
The input and output parameters will be trained using the ANN technique.The inputs, hidden, and output layers of the ANN levels are 1, the input layer unit is 4, the hidden layer and output layer unit is 5, and Epochs are 1,000.The back propagation neural network algorithm (gradient descent learning rule) was used for processing ANN.The sigmoid activation function was used to train the neural network model.
Figure 5 shows the machining time ANN structure, and Figure 6 depicts the input, hidden, and output layers used in the ANN model.The machining time was estimated using the NN trainer, which drives through 1,000 repetitions, as shown in Figure 7. Figures 8 and 9 show the regression plot and ANN validation performance; these show the gradient, validation check, and learning rate.Figure 10 shows the  The output of experimental and trained ANN data is shown in Figure 11.This graph shows the experimental results obtained by using ANN are very close, and evaluations are accurate to their values.The output of experimental, RSM, and trained ANN data are shown in Figure 12.This graph shows that the experimental, RSM result obtained from Design Expert software and ANN result are adjacent to their output values.Figure 13 shows the error percentage between RSM and ANN over the experimental value.The ANN predicted values show the lowest deviation than RSM values.The ANN was successfully trained for better prediction than RSM.

Validation of the Model Using ANN Transfer Functions
Various types of standardized statistical performance evaluation criteria are used to assess and validate the models applied in ANNs.Three different kinds of transfer functions have been used for neurons in hidden layers: hyperbolic tangent sigmoid (TANSIG), log sigmoid (LOGSIG), and PURELIN are compared and investigated for improving the performance of the proposed neural networks [32][33][34].

Mathematical Problems in Engineering
The ANN is trained by transfer functions using scaled gradient Descent with Momentum & Adaptive L.R. method.TANSIG and purely transfer function models are trained in the ANN network to compare whether the model is efficient and provides less predicted value.The 15 datasets are selected for testing.The results are compared; the LOGSIG is the most suitable transfer function that reflects the minimum prediction result.The predicted value of the three transfer functions is shown in Table 5.The validation result is shown in Table 6.

Validation Using Gray Relational Analysis (GRA)
GRA is an efficient approach for maximizing the complex interrelationships between multiple performance variables.Prasad  8 Mathematical Problems in Engineering et al. [35] proposed the GRA technique to optimize the pulsed current MPAW process, considering quality factors.Srivastava et al. [36] utilized a combination of RSM and GRA approaches to improve the process parameters of the FDM.GRA was used to handle the multiresponse optimization problem by altering the weights for distinct replies based on the quality or productivity needs of the process [37].Kirti and Raju [38] studied how to use GRA to optimize the WEDM process parameters of Inconel 600 with other performance attributes.Deng [39] proposed GRA based on the gray system theory to solve the intricate interrelationships among several responses.The objective is to reduce machining time.The procedures are mentioned as follows: (1) The experimental values were first normalized from 0 to 1. Mathematical Problems in Engineering (A) Smaller-the-better criteria (B) Higher-the-better criteria where y i (m) is the value after gray relational generation; min x i (m) is the smallest value of x i (m) for the mth response; max x i (m) is the largest value of x i (k) for the mth response.(2) The gray relational coefficient (γ) was computed using normalized experimental findings in the next phase.
where γ i is the distinguishing coefficient: if γ is smaller, then the distinguishing ability is larger.In common, γ = 0.5 is used; Δ min represents the smallest value of ΔO i (m); Δ max represents the largest value of ΔO i (m), ε represents the distinguishing coefficient.(3) The overall gray relational grade (GRG) was calculated by averaging the gray relational coefficients for each chosen response.
where X i represents the overall GRG; w k represents the normalized weightage of factor m.
The relevance of factors with uneven weightage is borne by diverse aspects of a simple engineering system.
The primary purpose of this research is to minimize machining time.As a result, GRA optimization, the "lower is better" approach, was applied.To begin, trial data were standardized using Equation (3) to achieve gray relational generation, as tabulated in Table 7.The GRA coefficient (γ) was computed from normalized experimental results using Equation (5).The overall GRA grade (X i ) was calculated by averaging the GRA coefficients for each selected response.For all of the processing, variables are weighted;  Mathematical Problems in Engineering equally, X i equals 0.5.If all of the processing variables are weighted equally, X i equals 0.5.Table 5 shows the overall gray relational rank.As a result, the level with the highest GRA grade value is the optimal level for the machining variables.

Conclusion
The turning operation was performed using DSS 2025 workpiece material and input factors: cutting velocity, rate of feed, cutting depth, and tooltip nose radius.The 2nd-order statistical model is developed to forecast machining time utilizing CCD and DoE of RSM.This method examines the interaction impact of the process parameter.A heuristic ANN is used to optimize and analyze the ANN model, which yields more accurate forecast data.In terms of prediction, the findings demonstrate that the ANN outperforms the RSM.
The cutting velocity has the most significant influence compared to the other parameters, and the machining time is low when the cutting velocity of 100-140 m/min.

Mathematical Problems in Engineering
To produce the optimum results, the nose radius of the cutting tool should be 2.8 mm.In the ANN optimization, the observed values and the ANN projected values correlate closely.
The results are further evaluated using GRA and exposed that the ideal machining settings are V c = 140 m/min, F z = 0.5 mm/rev, D c = 0.5 mm, and R tip = 2.4 mm.
The ANN and GRA results are the same; it is confirmed that ANN performs well.Confirmation tests were done using the GRA method and were validated for the performance measures.The anticipated data from the ANN has a margin of error of less.Consequently, the model is acceptable, and the optimum cutting parameters ensured significant

Future Studies
Studies and experiments in the fields of machinability, impact toughness, MRR, etc., will be carried out in order to optimize the DSS 2025 machining process.To ascertain the effects of cutting fluids, MQL, and cryogenic machining, further research must be done.Modifying the nomenclature of the tools to guarantee precise machining.Future use of sustainable machining technology is possible to reduce environmental pollution.
u t t i n g v e l o c i t y ( m / m i n ) B : R a t e o f f e e d ( m m / r e v )

FIGURE 1 :
FIGURE 1: 3D plots (V c versus F z ) over T m .

FIGURE 2 :
FIGURE 2: 3D plots (V c versus D c ) over T m .

FIGURE 3 :
FIGURE 3: 3D plots (V c versus R tip ) over T m .

FIGURE 5 :
FIGURE 5: ANN structure for machining time.

TABLE 1 :
Response table of process parameters.

TABLE 3 :
Analysis of variance to predict-T m .

TABLE 4 :
Percentage contribution of ANOVA.

TABLE 6 :
Validation of ANN data transfer functions.machining time.The result of the experiment assists the industrialists in improving the performance measure by reducing the machining time, which saves the operator energy/cost and improves productivity.

TABLE 7 :
Gray relational generations-analysis of turning data.