A new generation of hybrid nanofluids (HNFs) has been created as a result of emerging technologies [1]. Unlike nanofluids (NFs), which only contain one metal nanoparticle, HNFs contain many metallic nanoparticles. A simple NF is created when water is added to A l 2 O 3 nanoparticles [2]. However, combining water, aluminum oxide, silicon oil, and copper metallic nanoparticles results in the formation of a HNF. The thermal conductivity of HNFs is much higher than that of the common fluids, which is one of the reasons behind the interest of numerous scientists and academicians in this area. Nadeem et al. [3] analyzed the important properties like transfer of heat for the 3D HNF flow. Suresh et al. [4] studied the hybrid NF’s mobility and heat transport features. Gorla et al. [5] employed the source or sink impact on the Cu- A l 2 O 3 / H 2 O to explore natural convection in conjunction with heat transmission. Once they moved the heat sources to achieve hybrid prohibition, the Nusselt number decreased dramatically. Tayebi et al. [6] presented a comparative study of the HNFs Cu- A l 2 O 3 /water and NFs A l 2 O 3 /water through monocular elliptic cylinders by taking the convective effects. Acharya [7] studied the radiative natural convective HNF flow patterns and thermal transport through a square object with variable heated walls. The heat transmission of a HNF through the pores of a cylindrical shape is covered by Tayebi et al. [8]. Using a rotating apparatus, Chamkha et al. [9] analyzed the magnetohydrodynamic HNF flow for heat transfer. Asghar et al. [10] studied the ooze slime for examining the gliding motility of bacteria. The kinetic control of the rod-like swimmers past a sinusoidal object is explained by Asghar et al. [11]. A more recent survey on micro-swimmers and bacteria can be found in [12–16]. Acharya [17] considered the particle size and the solid–liquid interface by investigating the magnetic field effects for ferrofluid flow through a rotating disk. Yarmand et al. [18] studied the accelerated heat transport for the HNF flow. Abbasi et al. [19] investigated the thermal expansion coefficient for HNF of alumina and carbon nanotubes. Both theoretical and experimental studies for the thermophysical characteristics of the HNF flow through nanotubes are presented by Sajid et al. [20]. Kada et al. [21] studied the Carreau–Yasuda model for the waves of cilia in a micro-channel. Asghar et al. [22] used the numerical approach for simulating the flow in a micro-channel using the Carreau–Yasuda layer.

Rotating machinery is a vital component of numerous industries. The rotating disk has many industrial applications, and therefore, its analysis is very important. An increasing variety of industries, including the aviation, automotive, and marine sectors, are using these rotating objects in various parts [23, 24]. The use of irregularly thickened disks is growing, mostly due to financial limitations and the requirement to improve mechanical properties. Rotating disks are necessary for the operation of many pieces of industrial equipment. As a result, researchers working on this subject have recently launched a number of initiatives. Shah et al. [25] investigated the Hall current for the 3D NF flow extending surface by considering the Cattaneo–Christov (C.C.) heat flux model. For example, the analysis of NF flow resulting from rotating disks with different thicknesses and relatively uniform responses was done by Hayat et al. [26]. In a different work, Hayat et al. [27] investigated the flow between two stretchable rotating disks in a porous medium using the C.C. heat flux simulation. The analysis of heat and mass for the 3D NF flow past an elastic sheet is investigated by Khan et al. [28]. Qayyum et al. [29] studied the entropy and dissipation of the MHD Williamson fluid between two rotating disks. Jyothi et al.’s investigation deals with the effect that magnetic fields and heat radiation have on CNT convection in NF within rotating elastic disks [30]. Pourmehran et al. [31] used the Patel model and Brownian motion to study heat exchange and NF flow between two rotating disks. A more recent work on rotating surfaces can be found in [32–35].

The applications of artificial intelligence (AI) cannot be denied. AI covered all areas of research, including medical, engineering, and technology [36–40]. The engineering applications of AI are briefly reviewed by Nti et al. [41]. Jang et al. studied the AI applications for the recognition of pathways and enzymes in metabolic engineering. Sofos et al. [42] reviewed the applications of AI in the field of fluid mechanics. Kartik et al. [43] analyzed the inviscid flow field by using AI. Amini and Mohaghegh [44] used AI by analyzing the fluid flow in a porous medium. The squeezing model with the help of AI is analyzed by Almalki et al. [45]. The irreversibility impact of considering the carbon nanotubes during the viscous fluid flow is analyzed by Zubair et al. [46]. They used a supervised learning-based AI approach to analyze the fluid flow. The MHD HNF flow during the rotating frame for heat and mass transfer is investigated by Shoaib et al. [47]. They used the numerical result as a reference solution for the neural network and analyzed the problem for various impacts of the pertinent parameters and related statistical analysis. The thermal slip and absorption impacts on the HNF flow during the rotating disk are investigated by Shoaib et al. [48]. They used AI for computational purposes. Ali et al. [49] used the Levenberg–Marquardt backpropagation search path for training the neural network for the analysis of the water-based CNT HNF fluid past an unstable spinning disk.

The magnetization of conducting fluids is the subject of research in the field known as magnetohydrodynamics (MHD). MHD engagement detects the interaction of ferromagnetic or fluid metal particles in the presence of an electromagnetic field and current. The MHD model connects the electrohydrodynamic Maxwell equations and fluid computations with the Lorentz force as a result of magnetism. The processes that generate Lorentz force and capacitive electric charge generally seem to be completely opposite to one another. Due to a decrease in concentration, there is an increase in temperature, growing velocity, and Joule heating as the joule number increases. Babu and Sandeep [50] studied the 3D slip effects for the MHD NF flow past an unstable sheet. Ghadikolaei et al. [51] studied the MHD stagnant-point flow of a T i O 2 –Cu–water hybrid NF across such a stretched sheet. The T i O 2 nanoparticles in a porous medium were studied in the boundary layer of a micropolar hazy fluid medium by Ghadikolaei et al. [52] in order to take into consideration the effects of magnetization and heat radiation. The flow of thin fluid films with different thermophysical properties in a two-dimensional boundary layer in a three-dimensional environment was studied by Palwasha et al. [53].In light of the above, we assume the following.

• The HNF (Ag+ T i O 2 /silicon oil) flow past a spinning disk is considered.

Assume a steady HNF flow past a spinning disk that is axially symmetric along the φ direction in the ( r , φ , z ) coordinates. The disk is half-filled with HNF and rotates with angular velocity Ω . The surface and the surrounding temperatures are T w and T ∞ , respectively. A magnetic field B is acting normal to the z -direction. Since the flow is axially symmetric, the implications along φ are not considered, as shown in Figure 2.

Considering the Buongiorno model with modifications, we have the following basic equations for the fluid flow [54, 55]: u r + u r + w z = 0 , (1) ρ h n f u u r − υ 2 r + w u z = − P r + μ h n f u r r + 1 r u r − u r 2 + u z z − σ h n f B 0 2 u − μ h n f k * u − F u 2 , (2) ρ h n f u u r − u υ r + w υ z = μ h n f υ r r + 1 r υ r − υ r 2 + υ z z − σ h n f B 0 2 v − μ h n f k * υ − F υ 2 , (3) ρ h n f u w r + w w r = − P z + μ h n f w r r + 1 r w r + w z z , (4) u T r + w T z = α h n f T r r + 1 r T r + T z z + τ [ D B T r C r + T z C z + D T T ∞ T r 2 + T z 2 + 16 σ † T ∞ 3 3 ρ c p h n f k T z z , (5) u C r + w C z = D B C r r + 1 r C r + C z z + D T T ∞ T r r + 1 r T r + T z z − K r 2 T T ∞ n ⁡ exp − E a k T C − C ∞ , (6) with the B.Cs [56]. at z = 0 : w = 0 , u = L 1 u z , υ = r Ω , C = C w , T = T w , as z → ∞ : T → T ∞ , u → 0 , C → C ∞ , P → P ∞ . (7) Here, u , v , and w are the velocity components along r , φ , and z directions, respectively. In addition, υ f , T , D T , D B , C , P , α h n f , τ , ρ f , k * , F , L 1 are the kinematic viscosity of the fluid, fluid temperature, thermophoretic and Brownian diffusion coefficients, nanoparticle concentration, pressure, thermal diffusivity of the hybrid NF, the ratio of specific heat to the nanoparticle element, density, porous medium permeability, porous medium inertial coefficient, and the wall slip parameter, respectively.Introducing the following dimensionless transformations [31, 55]: P , C , T = P ∞ − Ω μ f P η , C ∞ + C w − C ∞ ϕ η , T ∞ + T w − T ∞ θ η , u , w , υ = Ω r f ′ η , − 2 Ω υ f f η , Ω r g η , η = 2 Ω v f z . (8) Applying Eq. 8 to Eqs 1–6, we obtain [55]: f ′ ′ ′ + ρ hnf ρ f μ hnf μ f f ′ ′ f − 1 2 f ′ 2 + 1 2 g 2 − ρ hnf ρ f μ hnf μ f f ′ 2 F r − δ f ′ − M σ hnf σ f μ hnf μ f f ′ = 0 , (9) g ′ ′ ′ − ρ hnf ρ f μ hnf μ f 1 2 f ′ 2 f − 1 2 f ′ g − f g ′ − M σ hnf σ f μ hnf μ f g − ρ hnf ρ f μ hnf μ f g 2 F r − δ g = 0 , (10) 1 P r k h n f k f ρ C p hnf ρ C p f + 4 3 R d θ ′ ′ + θ ′ f + N t θ ′ 2 + θ ′ N b ϕ ′ = 0 , (11) ϕ ′ ′ + S c f ϕ ′ + N t N b θ ′ ′ − σ 1 S c 1 + δ ̃ θ n ⁡ exp − E 1 + δ ̃ θ ϕ = 0 , (12)  as  η = 0 : f ′ = α f ′ ′ , g = 1 , f = 0 , θ = 1 , ϕ = 1 ,  as  η → ∞ : f ′ → 0 , g → 0 , f = 0 , θ → 0 , ϕ → 0 . (13) Here, R d = 4 σ T ∞ 3 k k f is the radiation parameter, M = σ f B 0 2 2 ρ f Ω is the magnetic parameter, S c = v f D B is the Schmidt number, N b = τ D B C ∞ v f is the Brownian motion parameter, E = E a k T ∞ is the energy activation parameter, F r = C b k * represents the DF number, δ ̃ = T w − T ∞ T ∞ is the temperature difference parameter, α = L 1 2 Ω v f is the velocity slip parameter, P r = v f α f is the Prandtl number, σ 1 = K r 2 Ω is the chemical reaction parameter, δ = v f k * Ω is the porosity parameter, and N t = τ D T T w − T ∞ T ∞ v f is the thermophoresis parameter.It is clear from Eq. 4 that we can obtain the pressure by integrating it. The torque resistance on the disk having radius R is given as follows [55]: T = − ∫ 0 R μ h n f v z z = 0 2 π r 2 d r = − π ρ h n f Ω 2 2 υ f Ω R 4 g ′ 0 . (14) In addition, the tangential τ 0 and radial τ r stresses are given as follows [55]: τ r = μ h n f u z + w r z = 0 = r Ω μ h n f 2 Ω υ f f ′ ′ 0 , τ θ = μ h n f υ z + w r z = 0 = r Ω μ h n f 2 Ω υ f g ′ 0 , (15) or C f = τ r 2 + τ θ 2 ρ h n f r Ω 2 = 2 Ω r 2 υ f − 1 2 f ′ ′ 0 2 + g ′ 0 2 1 2 1 − ϕ 1 2.5 1 − ϕ 2 2.5 . (16) The local Nusselt number N u r is computed as follows [55]: N u r = r q ′ ′ k f T w − T ∞ , (17) where q ′ ′ = − k hnf  T z + 16 σ † T ∞ 3 3 k T z z = 0 is the heat flux. The wall heat flux can be eliminated by choosing the zero wall mass flux; thus, with Eq. 8, we have from (17) N u r = − r 2 Ω υ f k h n f k f + 4 3 R d θ ′ 0 . (18) The HNF models and physical properties of the NFs used in this study are displayed in Tables 1,2.

Artificial neural networks (ANNs) are computer models that simulate the human brain structure. The human system, from a neurological point of view, is very complex. ANNs are made up of nodes, which are interconnected to form layers with varying degrees of processing depth. Nodes are connected processing components. These combine to create intricate processing circuits that identify input patterns and generate responses. By training layers that correlate responses with incoming data with specific pattern knowledge, the patterns can be learned. The network gains knowledge through practice, and when linked to other networks, it gains knowledge through information sharing. The only layers in an ANN’s structure are the input layer, a hidden layer that is introduced first, and the output layer. Applications of neural networks are being used across numerous industries to address issues with security, the economy, and other factors. In the data-intensive era, neural networks have created new opportunities for study and application [60, 61]. The more recent trends in other disciplines of AI can be found in [61–63]. For a solution of Eqs 9–13, we first transform the given system into a first-order system. For this, we have [55] y 1 = f , y 2 = f ′ , y 6 = θ , y 7 = θ ′ , y 3 = f ′ ′ , y 4 = g , y 5 = g ′ , y 8 = ϕ , y 9 = ϕ ′ y 1 ′ = y 2 , y 2 ′ = y 3 , y 3 ′ = − A 0 / A 1 y 1 y 3 − 0.5 y 2 2 + 0.5 y 4 2 − F r y 2 2 + δ y 2 + A 2 / A 1 M y 2 , y 4 ′ = y 5 , y 5 ′ = A 0 / A 1 0.5 y 2 2 y 1 − 0.5 y 2 y 4 − y 1 y 5 + F r y 4 2 + A 2 / A 1 M y 4 + δ y 4 , y 6 ′ = y 7 , y 7 ′ = − P r y 1 y 7 + N b y 7 y 9 + N t y 7 2 / A 3 / A 4 + 4 / 3 R d , y 8 ′ = y 9 , y 9 ′ = − S c y 1 y 9 + N t / N b y 7 ′ − σ 1 S c 1 + δ 1 y 6 m y 8 e x p − E / 1 + δ 1 y 6 (19) The B.Cs are as follows [55]: y 1 0 = 0 , y 2 0 − α y 3 0 = 0 , y 4 0 − 1 = 0 , y 6 0 − 1 = 0 , y 8 0 − 1 = 0 , y 2 ∞ = 0 , y 6 ∞ = 0 , y 4 ∞ = 0 , y 8 ∞ = 0 . (20) The two important steps of ANN are shown in Figure 3. During the process, after providing the input, the weights are trained for t k inputs, and the required result is obtained as an output. Mathematically, we have [64, 65] u j = ∑ j = 1 k w j t j − β j . (21) The results for the state variables can be obtained by introducing the sigmoid function χ ( η ) = 1 1 + e − η . The obtained output is given by [66] χ u j = 1 1 + e − w j t j − β j . (22)

The impact of the silicon oil-based HNF ( T i O 2 +Ag) flow past a rotating disk using the Buongiorno modified model has been analyzed with the trained AI method. In this section, the results are given under the influence of various pertinent parameters. For simplicity, we choose 0 ≤ M ≤ 3 , P r = 6.2 , 0 ≤ δ ≤ 1,0 ≤ R d ≤ 3,0 ≤ N t ≤ 1,0 ≤ N b ≤ 2,0 ≤ E ≤ 2 , δ ̃ = 1,0 ≤ F r ≤ 5 , S c = 1 , 0 ≤ α ≤ 1 , and n = 3 .

The effect of various parameters on the radial and axial velocities is described in Figure 4A–H. The larger values of the velocity slip parameter α increase both the axial and the radial velocities, as depicted in Figures 4A,B. The variations are very prominent for the HNFs, as compared to the NFs. The HNFs and the NFs behave the same after η = 3 . Physically, the greater values of α shift the fluid toward the rotating wall of the disk, which further enhances the fluid flow in both the axial and radial directions. In addition, α is directly related to the spinning factor, which plays a key role in the increasing trends of the fluid flow. The porosity parameter δ impact is displayed in Figures 4C,D. The larger values of δ decrease both the axial and radial velocities. The larger values of δ increase the kinematic viscosity and further decline the rotational effects, which, as a result, causes a decline in both profiles. In addition, the larger values of δ decrease the rotational velocity of the surface that plays a key role in the decline of the velocity profiles. These variations are much faster for HNFs than those for NFs. The impact of M and the DF parameter F r is displayed in Figure 4E–H. Both profiles decline with the increasing values of these parameters. These variations are due to the permeability parameter that acts oppositely to the larger trends of F r . On the other hand, M produces the Lorentz force that acts in the opposite direction to the fluid flow and, as a result, decreases the velocity profile, as shown in Figures 4G,H. When M increases, the rotational velocity of the disk declines, which further declines the axial and radial velocities.

The absolute errors (AEs) to minimize the L 2 − norm are displayed for all the variations in the above-mentioned parameters in Figure 5A–H. The axial velocity AEs lie in the range 1 0 − 4 to 1 0 − 8 , while the radial velocity lies in the range 1 0 − 3 to 1 0 − 9 . The best and minimum occur in both variations for α = 0.9 at η = 2.4 and η = 2.2 , respectively, as depicted in Figure 5A,B. The AEs for δ and F r are displayed in Figure 5C–F. The range for both variations is 1 0 − 4 to 1 0 − 9 . The minimum threshold for both parameters is obtained at δ = 0.5 and F r = 3.0 . The axial velocity AE varies in the range of 1 0 − 4 to 1 0 − 10 for M , while the radial velocity ranges from 1 0 − 4 to 1 0 − 9 . The minimum value achieved for the radial and axial velocities was at M = 0.5 and M = 1 , respectively.The error histograms for each parameter with its varying values are presented in 6(a)–(h). The bar graphs are presented with the errors vs the instances with the minimum bins at which the required threshold is achieved. The zero error occurs at the midpoint of the range − 3.8 e − 6 - 2.52 e − 6 for the radial velocity, while the same result occurs at 2.3 e − 6 for the axial velocity with the varying values of α . The zero error for 4,000 and 8,000 instances is depicted in Figures 6C,D at errors − 3.3 e − 6 and − 6.9 e − 6 for the radial and axial velocities with varying values of δ , respectively. The Darcy’s parameter impact on the radial velocity shows zero error at the midpoint of − 7.6 e − 6 - 5.09 e − 6 for 3,500 instances. This result is achieved at 20 bins, as shown in Figure 6E. This result for the axial velocity with the varying values of F r is obtained at − 3.19 e − 6 with 6,000 instances, as depicted in Figure 6F. The effect of M on both the velocities is displayed through error histograms in Figure 6G,H. For the radial velocity, the zero error occurs at the midpoint of − 1.2 e − 5 - 6.19 e − 6 , while for the axial velocity, the same result is obtained at − 4.7 e − 6 with greater M .

The effect of α for both radial and axial velocities is displayed through the regression lines in Figure 7A–D. The regression lines are used for the fitted data that are displayed with a linear line. Here, the target is taken at the x -axis, while the output is considered at the axis perpendicular to it. The first three horizontal boxes in row one from left to right show the fitted data for the NFs with α = 0.1 , 0.5 , 0.9 . The value of R in each case is 1, which shows that 100 % of the data is fitted. The same results are observed for the HNFs in row two, where R = 1 , as shown in Figure 7A, which is the collective data fitting for the radial and axial velocities with varying values of α . The regression lines for the porosity parameter δ , Darcy’s parameter F r , and magnetic parameter M are displayed in Figure 7B–D. The results show that 100 % data are fitted for the total data considered, as shown with hollow circles. The output range for the HNFs is smaller than the NF range. The fitted data are displayed with various colors (blue, red, green, and black) displayed with circles on the linear line. The minimum output is obtained for the magnetic parameter M for the HNFs.

The validations for the total performance of each parameter are presented in Figure 8A–D. These figures demonstrate the mean square error (ms) vs the total iterations performed. Four different curves are used for the trained, validated, tested, and best data. The performance for α and δ is displayed in Figure 8A,B. The mean square error at the best point occurs near 1 0 − 10 displayed with the green circle. This point is achieved at 3.024 e − 10 and 3.2758 e − 10 for both α and δ , respectively. In addition, the performance for F r and M is displayed in Figure 8C,D. The best performance is achieved at 9.2472 e − 11 and 1.4931 e − 09 for these two parameters, respectively.

The impact of N b , N t , R d , and E is analyzed on the thermal and concentration profiles in Figure 9A–F. The varying values of N b on the thermal and concentration profiles are described in Figure 9A,B. The increasing values of N b increase both profiles and vary in the range of 0–1. Physically, when N b increases, the kinematic viscosity decreases and the ambient concentration jumps, which, as a result, increases the number of suspended particles. When these particles increase, the interaction between them also becomes faster and more random, which, as a result, enhances energy transfer from one point to another. These interactions play a key role in energy transfer, either by convection or conduction. The thermophoresis parameter effect shows an opposite trend for both profiles, as shown in Figure 9C,D. The larger values of N t also increase the temperature difference and reduce the kinematic viscosity. The larger the N t , the higher the difference in temperatures, and the greater the thermal profile, as depicted in Figure 9C. The concentration profile falls due to increase in the density of the fluid with larger values of N t . Furthermore, the diffusion coefficient increases with larger values of N t , but this phenomenon occurs at a very small scale that cannot interrupt the profile mechanism, as displayed in Figure 9D. Both the radiation parameter and the energy activation parameter enhance the thermal and concentration profiles, respectively, as shown in Figure 9E,F. The larger values of R d increase the inner temperature due to an increase in the ambient temperature. The activation energy also enhances the concentration profile, as shown in Figure 9D.

The corresponding mean square error for the varying values of N b , N t , R d , and E is depicted in Figure 10A–F. It is clear from Figure 10A,B that when N b varies from 0.5 to 1.5, the corresponding L − norm achieves its minimum values of 1 0 − 8 and 1 0 − 9 for θ and ϕ , respectively. A very similar trend for the errors for their minimal values occurs in Figure 10C–F for the remaining parameters. The ms for the activation of the energy parameter E is approaching 1 0 − 10 . This shows the total performance of the impact of E on the concentration profile with its increasing values, as depicted in Figure 10F.

The error histograms for all these parameters ( N b , N t , R d , E ) are displayed in Figure 11A–F. The zero error for N b of θ occurs at − 7.6 e − 6 , while for the concentration profile, it occurs in the range − 1.5 e − 5 – 6.85 e − 6 . All these results are obtained at 20 bins by using 6,000 and 2000 instances, respectively. The zero error for N t by computing the thermal and concentration profiles falls in the range − 7.1 e − 6 – 9.7 e − 6 and − 2.4 e − 5 – 2.82 e − 5 , respectively, as presented in Figure 11C,D. In addition, for the impact of the radiation and the activation parameters, the zero errors are obtained at 5.08 e − 6 and in the range of − 2.1 e − 5 – 2.97 e − 5 , respectively.

The regression lines for the impact of the parameters ( N b , N t , R d , E ) are dispayed in Figure 12A–D. The results for the increasing trends are presented for the NFs and HNFs separately in two rows. The output is presented on the y -axis, while the target points are chosen along the x -axis. In each case, the regression shows R = 1 , which proves that 100 % of the data is used in the fitting line. The minimum output for all the cases ranges up to 1 0 − 18 , which proves the accuracy of the performed analysis. The fitted data are displayed in various colors (red, blue, green, and black).

The validation performance for the impacts of N b , N t , R d , and E is displayed in Figure 13A–D. In Figure 13A,B, the performance for N b on θ and N t on ϕ is presented. The best and minimal mean square errors are achieved at 2.1447 e − 10 and 3.2913 e − 9 for the parameters, respectively, using 1,000 epochs. Similarly, for R d and E , the same results for the mean square error are achieved at 4.638 e − 10 and 4.1829 e − 9 , respectively.

This article covers the neural network applications of the silicon-oil based HNF flow past a spinning disk using the Buongiorno model modifications. The impacts of various pertinent parameters for the thermal, concentration, and the velocity profiles are briefly described. We conclude the following:

• The thermal profiles of nano and HNF increase with increasing values of R d , N t , M , and N b .