The aluminum alloy square plates (100 mm × 100 mm, 6 mm thick) with the designations AA2024-T351 and AA7075-T651 were used for experimental investigation. The FSW tools determine the mechanical, weld quality, and micro-structural characteristics of the FSWed joint (Cui, L et al., 2018). FSW tool used for experimentation is AISI H13 tool steel, hardened, and tempered. Because of its high toughness and fatigue resistance, H13 tool steel is perfect for both hot and cold applications. The welding tool for this experiment has been designed and manufactured to have a D/d ratio of 2.5, shoulder diameter of 20 mm, threaded pin diameter of 8 mm with length of 5.8 mm. The reinforcing material used in this investigation is nano alumina (Al₂O₃) with an average particle size of 33 nm and a constant 8% volume fraction. The reinforcement percent of nano alumina has been decided by the number of holes drilled in the FSW plates and volume fraction (Sadeghi, B., et al., 2022). The FSW process is performed in VDF1200 CNC Vertical Milling Center. 20 mm thick mild-steel plates have been in proper dimensional sizes to secure the two dissimilar aluminum plates in the butt position. A backing plate of 20 mm thickness was used underneath the aluminum to protect the milling table from frictional heat and to rigidly support the weld material against the axial force from the welding tool.

Fifteen equally spaced holes of diameter 2 mm and length 4 mm on each edge of both dissimilar plates of aluminum alloys AA 2024-T351 and AA7075-T651 have been drilled and cleaned with acetone to remove the possible grease, dirt, and stains effectively. A volume fraction of 8% Al₂O₃ Nano-powder was mixed with methanol (CH₃OH) of 99% purity, and the resulting thick slurry was charged manually into the 15 drilled lateral holes of both aluminum plates, as shown in Figure 1. The multi-pass FSW was carried out using a VDF- 1200 CNC Vertical Machining Center.

Response surface method (RSM) is an effective scientific method for modeling and analyzing problems when many variables influence the response of interest and where the goal is to optimize the parameter (Cui, L et al., 2018). The RSM approach was utilized to assess the relationships between the independent variables and the responses, predict responses, and complete the optimization in the intended range. The statistical model was generated for each response after the FSWed characteristics were measured, and the relationship between the parameters and the FSWed characteristics was assessed. The current study attempted to optimize process parameters by creating a quadratic (2nd order) function to characterize FSW process performance. The equation was then used to establish the process at its optimal setting, which could be achieved through the use of CCD. Furthermore, the CCD proves to be an effective tool for researching the relationships between process factors. The second-order RSM regression model links the process parameters and the responses. The form of the RSM model is shown in Eq. 1 as follows. y = β 0 + ∑ i = 1 k β i x i   + ∑ i = 1 k β i i x i 2 + ∑ i   ∑ j β i j x i (1)

Where: y is response, β₀ to β_ij are coefficients of unknown constant computed from the analysis, x_i are independent variables, k is the quantity of independent variables.

The study employed a Central Composite Design (CCD), the most commonly used and preferred experimental design for RSM (Bhattacharya, S., 2021). A CCD model may have the desired characteristics of rotatability and orthogonal blocks, depending on the alpha value to reduce the variation in the regression coefficients. The CCD’s alpha value (α) is influenced by the total number of experiment trials in the factorial section to preserve rotatability. This study assessed the impact of process variables on the tensile strength (TS), yield strength (YS), and microhardness (MH) of FSWed joints made between dissimilar grades of aluminum alloys, 2024-T351 and 7075-T651. The DoE and analysis of the FSW process are performed with design expert software. Table 1 shows the FSW factors and levels. The FSW parameter ranges were considered for current studies based on CNC machine centers with operating ranges and a comprehensive literature survey (Teja and Srinivasan, 2024). Twenty experimental factors and levels were used, and the experimental results of FSW of dissimilar aluminum alloys are shown in Table 2.

Material testing and characterization are crucial in evaluating the FSWed joints’ quality attributes. The study employed the Tensile Strength (MPa), Yield Strength (MPa), and Micro-hardness (HRB) tests for mechanical properties analysis. The Scanning Electron Microscope (SEM) and optical microscope were used to investigate the microstructure and topographical analysis.

A regression model describes the interaction among several independent factors and a dependent variable. Hence, the model formulates the equations to predict responses. The final model equations (Eqs 2–4) created predicts the UTS, YS, and Microhardness of the AA 2024-T351 and AA 7075-T651 DFSW joints. The statistical significance of any differences between the means of multiple groups is ascertained by applying Analysis of Variance (ANOVA).

Tensile strength, yield strength, and microhardness are three responses that are significantly affected by the interaction of process parameters. As a result, the CCD approach was chosen for process parameter optimization. CCD provides the mathematical model used for predicting the weld’s tensile strength, yield strength, and microhardness, shown in Eqs 2–4. In order to determine the significant factor, the ANOVA approach investigated the parametric effect. It establishes that responses are dependent on the parameters. Comparing the mean square against all experimental data, which yields the ANOVA test, resulted in Tables 3–5, examining each significant component’s importance and interaction. Insignificant factors are removed from ANOVA by a backward elimination process. The accuracy of the models created based on the data was assessed using an F-test of ANOVA, and the results are shown in Tables 3–5. According to the assessments, these values represented the regression model’s best fit at a 95% confidence level, suggesting that the statistical models used were accurate. Y T S = 148.53 + 3.54 A + 2.48 B + 3.75 C + 5.82   A B + 6.89 A C + 7.17 B C − 8.61 A 2 − 3.55   B 2 (2) Y Y S = 142.58 + 1.31 A + 1.71 B − 0.34 C + 4.78   A B + 3.69 A C + 4.36 B C − 8.28 A 2 − 5.20   B 2 − 2.14 C 2 (3) Y M H = 88.36 + 023 A + 2.64 B + 0.70 C − 1.19 A 2 − 2.38   B 2   − 0.33 C 2 (4)

Interaction between these three factors significantly affects tensile stress. Additionally, the tool rotational and welding speeds have quadratic effects corresponding to their significance. The yield strength is directly impacted by all three input factors, in order of significance, as shown by Eq. 3, additionally, interaction and quadratic effects are based on their relative relevance. As shown in Eq. 4, the linear effects of the following input variables and quadratic effects of all three factors significantly affect microhardness.

ANOVA determines whether the various factors significantly affect the response variables. According to the ANOVA Table 3, the main interaction and high-ordered model terms A² and B² significantly influence the tensile strength. The p-value from the ANOVA Table 3 is ≤ 0.05, showing a statistically significant model. Hence, this smaller p-value indicates a significant difference among the response variable levels. The second-order response surfaces are fitted accurately by the experimental values.

The tensile values increase while the tool rotation increases; the highest tensile value is 170 MPa, obtained at high levels of FSW conditions such as 1,200 rpm, 60 mm/min, and 0.4 mm. It is because of the significant grain refinement brought on by the FSW process’s sustained dynamic recrystallization and extreme plastic deformation. The lowest tensile value of 119.2 MPa was found at 663 rpm. The tool rotation controls the level of heat developed during the FSW process (Zhang Z. et al., 2014; Dawood et al., 2015). The heat developed in FSW influences the material mixture and deformation due to plasticity (Kisuka F. et al., 2024; Salih et al., 2023). Due to inadequate material mixing, Al₂O₃ clustered, and the link between Al₂O₃ and the matrix material was weak in the FSWed joints manufactured at tool rotation 700 rpm (Chen X G et al., 2009).

At medium-level welding conditions, the tensile value is 150 MPa. The 3D surface and contour plot graphs in Figures 2A–C illustrate the relationship between rotational speed-welding speed and plunge depth on the tensile strength. The 3D and contour plot (Figures 2A, B) shows that the tensile strength increases while increasing the rotational speed, and similarly, the welding speed and plunge depth increase the tensile strength, as shown in Figure 2C. The changes in tool penetration depth, welding speed, and tool rotating speed can also alter the frictional and stirring conditions, impacting the joint’s tensile strength (Vahdati M. et al., 2020). Additionally, the tensile strength improves as the welding speed increases because of adequate heating and material agitation (Sabari S. et al., 2016).

The FSWed joint formed at 1,100 rpm tool rotation, 40 mm/min traversal speed with 10% Al₂O₃ nanoparticles demonstrated the enhancement of mechanical performance resulting from microstructure refinement and production of strain-free refined grains during the DRX mechanism (Jain and Mishra, 2022b). Significant strain evolved at an improved RS of 1,100 rpm, which led to the observation of homogeneously dispersed Al₂O₃ and appropriate material mixing. The absence of Al₂O₃ clustering indicates excellent bonding between the matrix and Al₂O₃. Higher RS values improved particle mobility (Palanivel R et al., 2016). The pinning action of the Al₂O₃ particles enhances the nucleation sites and prevents grain formation (Srivastava M et al., 2019). The greater pinning action with a higher percentage of more evenly dispersed Al₂O₃ boosted grain refinement (Azizieh M et al., 2011).

The yield strength of the Al₂O₃ nanoparticle-reinforced dissimilar aluminum alloy FSWed joints was investigated. The main interaction term and high-ordered model terms A², B², and C² significantly influence the yield strength given in Table 4. The relationship between factors on yield strength is shown by the 3D surface and contour graphs in Figures 3A–C. The highest yield strength is obtained by modifying the values of the tool rotating speed and welding speed in the range of maximum level, as shown in Figure 3A. Figure 3A illustrates how to achieve the optimum yield strength by varying the tool rotating speed and welding speed within the maximum level range. Accordingly, the yield strength increases while the rotating speed increases from 800 to 1,200 rpm. Here, frictional heat is increased by increasing the tool’s spinning speed. Alternatively, the yield strength increases as a result of more effective material mixing in the joint seam due to the frictional heat produced by expanding the tool’s shoulder diameter and the workpiece’s contact surface. Both the combined effect of welding speed and tool plunge depth show a similar tendency, as shown in Figures 3B, C. The yield strength is more affected by the combined influence of traverse speed and tool speed (Bhojan et al., 2016). Thus, the yield strength will increase if the welding and rotating speed increase. In this case, frictional heat increases as the tool’s rotating speed increases (Vahdati, M., et al., 2020). In contrast, higher yield strength is produced because of the more compelling material mixing in the joint seam caused by the frictional heat generated by a higher plunge depth. The highest yield strength is obtained at high levels of FSW conditions such as 1,200 rpm, 60 mm/min, and 0.4 mm. The lowest yield strength, 117 MPa, was found at 663 rpm, 40 mm/min, and 0.3 mm, respectively.

The ANOVA table of hardness is shown in Table 5. The weld nugget zone’s hardness could potentially have been claimed to be caused by the dispersion and grain refinement of Al₂O₃ particles within the WNZ. The heat input during FSW was found to be associated with the refined grain size in the WNZ. The hardness values vary from 77.1 HRB to 89 HRB of the dissimilar aluminum alloys AA 2024-T351 and AA 7075-T651. Table 2 displays the mean micro-hardness value at the midpoint of the weld zone for different reinforced joints. The maximum hardness of 89 HRB was found at high levels of FSW conditions such as 1,200 rpm, 60 mm/min, and 0.4 mm. The lowest micro-hardness values, 77.1 HRB, were 1,000 rpm, 6.36 mm/min, and 0.3 mm. Figure 4 shows the variation of the resulting micro-hardness values Al₂O₃ Nanoparticle reinforced FSWed joints.

The welded joint’s microhardness is typically predicted to be higher than the base materials. Microhardness increases during the FSW process due to intense plastic deformation and grain refining. Research findings indicate that the microhardness values of FSWed joints of AA2024 and AA7075 may vary between 77 and 89 HRB, contingent upon the particular FSW parameters utilized. The microhardness values typically result from tool rotation and traverse speed. The FSW process’s increased plastic deformation and grain refining are the main causes. These parameters influence the degree of heat input, the subsequent microstructural alterations, and ultimately, the hardness of the joint is affected. The hardness enhancement due to nano Al₂O₃ has a high joint density and exhibited a greater tensile strength than the joint made with TiO₂ nanoparticles (Singh et al., 2020). Since nano-sized particles were embedded in the WNZ, there was a more noticeable increase in microhardness and wear characteristics compared to joints without reinforcement under the same processing conditions (Singh et al., 2020). Increased grain refinements by nano Al₂O₃ were responsible for improved mechanical properties (Rani et al., 2022).

Scanning Electron Microscope (SEM) photographs of the Al₂O₃ nanoparticle-reinforced FSWed joint’s fractured surface morphology are displayed in Figure 5. SEM is an imaging technique that images a material’s surface using an electron beam. SEM images reveal that the specimens’ fracture surfaces include a variety of characteristics, such as shear lips, pits, and cracks. More microvoids were observed in low-level FSW conditions, as shown in Figure 5A. Little cracks discovered in images are displayed in Figure 5B. FSWed joints created at the rotational speed of 1,200 rpm and welding speed of 60 mm/min, 0.4 mm plunge depth produce consistent dispersion with substantial fragmentation of particles obtained and also particles agglomeration to be absent. The closer combination of such input parameters also enhances the mechanical parameters like tensile strength, % elongation, and micro-hardness (Jain and Mishra, 2022b). A significant volume fraction of different intermetallic particles can be found in al-alloy. Al₂O₃-based intermetallics are inclusions and particles in brittle phases distributed throughout the Al-matrix. Once plastic deformation occurs, the voids are initiated or nucleated mainly at the grain boundaries due to the incompatibility of the Al-matrix and intermetallic particles with differing characteristics. These features make it easy to investigate the mechanisms underlying ductile fracture and damage evolution. The Al-alloy’s in situ tensile experiments demonstrate that, for ductile materials, dimple-dominant fracture typically happens as depicted in Figure 5C (Li and Fu (2019). The specimens were fractured ductile, as indicated by the dimples that predominated on the fracture surfaces (Figure 5C).

Figure 6 shows the optical microscope image of WNZ (a) lower input parameters, (b) medium input parameters, and (c) higher input parameters. According to the microscopic images, the size of the grain particles is fine at medium and higher levels. The homogenous distribution of the Al₂O₃ nanoparticle is better when compared with low-level conditions. Adding nanoparticles to aluminum alloys AA2024 and AA7075 during FSW can refine grain, enhancing the joint’s strength and ductility. Similar results showed that the scattered SiC particles in the WNZ had a zener pinning effect and dynamic recrystallization, which refined the grains (Moustafa E B. et al., 2023). Results showed a pinning effect of scattered SiC particles in the WNZ and DRX, which led to grain refinement, causing an impact on the micro-hardness and tensile strength. Accordingly, higher ultimate tensile strength and maximum microhardness were recorded using 1,100 rpm tool rotation; 40 mm/min weld speed, and a SiC particle of 10% (Jain and Mishra, 2022b). Potential factors contributing to these changes include grain modification, ceramic particles in the interfaces, high applied strain that modifies the atomic network parameters very little, and the lamellar structure of the treated Al 1,050 in higher passes (Tayyebi et al., 2019). The primary distinction between raw material and processed Al layers is that dimples with the ARB process are smaller and shallower. This pattern has also been observed in prior comparable studies, which have found that using ARB techniques does not modify the failure mechanism for different Al alloys but rather affects the shape, quantity, and size of dimples (Rahmatabadi, D et al., 2019).

Figures 7A–C show the SEM images of surface topography at different FSW conditions with WNZ and TMAZ of the advancing and retracting sides of the weld joint. The welded zone displays the vortex flow pattern, and the mixed flow region consists of the AA2024 and AA7075 lamellae structures. The two alloys mix and penetrate deeply in this area, causing the plasticized material to experience extreme superplastic flow and turbulent intercalation patterns. Both alloys experience intense plastic deformation and in-situ extrusion due to the stirring tool action. The plasticized material is transferred layer by layer to form a lamellae structure. Especially at higher WS, the mixing is much more successful when the AA2024 is positioned on the AS, as shown in Figure 7C.

According to Myers et al. (2016), the desirability function approach is famous for multiple response optimization processes. This procedure finds the highest suitable response values under operating conditions x. This method transforms all the responses into distinct DFs (di) with a scale factor ranging from 0 to 1. If the desirability value is zero, the predicted value is deemed entirely undesirable, and the value is one denoted by a more desirable. These functions are organized by defining the values from the experiments as the desired outcome based on objective function minimization or maximization of response (y_i) (Derringer and Suich, 1980). Equation 5 determines each variable’s composite desirability (D) after the individual desirability has been determined. The predicted responses are weighted as one response considered maximizing. Individual desirability (di) is then determined using Eq. 6. D = d 1 , d 2 , d 3 … … … … … . d n 1 n (5)

Individual desirability for maximization as: d i = 0 y i < L y i − L T − L r L ≤ y i ≤ T 1 y i > T   (6)

The optimum parameters for the proposed desirability function are a tool rotation at 1,146 rpm, a transverse speed at 60 mm/min, and a plunge at 0.4 mm given in Table 6. These combinations produced a maximum performance in yield strength of 143.69 MPa, tensile strength of 165.65 MPa, and a hardness value of 88.5 HRB. The confirmation experiments are conducted for the predicted parameter values. The predicted optimal parameters produced joints gave tensile strength of 169 MPa, yield strength of 145 MPa, hardness of 89 HRB, and error values of less than 5% at a confidence level of 95%.

ANN models have been widely used as predictive strategies because of their affordability, relative ease of understanding, and capacity to be trained utilizing data gathered from these intricate manufacturing processes (Essa, A. R et al., 2023; Pany et al., 2001). An ANN is made up of three layers: the input layer, the hidden layer, and the output layer, as shown in Figure 8. The input layer contains all of the input parameters. The input layer’s data is processed by the hidden layer, and the subsequent output layer computes the subsequent output vector. The basic sequential computation elements of an ANN are modeled from the biological nervous system. An ANN’s neuron is its fundamental building block. Synapses, which connect neurons, are linked by a weight factor that is tied to each synapse. Using a single hidden layer, the Levenberg-Marquardt algorithm was applied in this study. Therefore, the network topology was composed of three neuronal outputs with a linear activation function, a logistic sigmoid, ten hidden neurons with a nonlinear activation function, and three input neurons. The 14 sets of data were used for training, and each three sets of data were used for the testing and validation process. Figures 9A–D illustrates predicted values are well agree with the measured data. Based on R-values 0.9991, 0.9754, 0.9846 for training, testing and validation respectively, the results demonstrate a high degree of similarity between the measured and predicted data sets. The ANN predicted, and experimental values are given in Table 6.