Prediction of cutting performance using artificial neural network during buffered impact damper-assisted boring process

In manufacturing industry tool vibration, tool wear and surface finish are the factors that affects the product quality and its production costs. During boring process, overhanging length of the tool holder generates tool vibration leading to poor surface finish, hastened tool life, and further reduction in machine tool life. For enhancing the cutting performance, Buffered Impact Dampers (BID) were designed, developed and tested in this research work. A set of 27-run cutting experiments was performed by varying particle size, material and filling. From the experimental results, Buffered Impact Dampers (BID) increases the rigidity of the tool holder which enhances the cutting performance. The particles in the boring tool will collide with one another thereby suppressing the tool vibration efficiently and enhancing the cutting performance when particle material is stainless steel, particle size Φ4 and particle filling is 75%. Artificial Neural Network (ANN) was implemented to predict the impact of buffered damper on surface roughness, tool wear, tool vibration and cutting force. The results obtained from ANN model were compared with the experimental results for MSE, AAD, MAPE and R. When comparing experimental results with ANN model, both the results concurred with each other.


Introduction
The dynamic interaction between workpiece and the cutting tool gives rise to tool vibration in metal cutting.The factors such as cutting force and surface finish are adversely affected by tool vibration, as a result of the slenderness ratio of the tool holder.In addition to that high-pitched, unpleasant noise, excessive tool wear and the tool vibration also weakens the machining system which causes damage to the machine parts (Lawrance et al. 2019).The length-to-diameter is the ratio of the length of the boring bar to its diameter.Rao et al. and Adarsha Kumar et al. (Venkata et al. 2015;Kumar et al. 2019) have found that a certain lengthto-diameter ratio can minimize tool vibrations and improve the performance of the boring process.It was identified that the best length-to-diameter ratio is 3 for less vibration in the boring process is dependent on various factors such as material being machined, cutting speed, feed rate, and other machining parameters.The passive dampers are frequently integrated in the metal cutting industry due to its low cost and efficiency to reduce the detrimental effects of tool vibration.Some commonly available passive dampers are Impact dampers (Lawrance et al. 2017) and Particle dampers (Lawrance et al. 2021).The particle dampers are a type of passive damper made up of a sealed container and a number of mass.When a tool vibrates, its momentum is transferred to the particles, the non-perfect elastic collisions and frictional losses in particles dissipate the energy, in turn dampening the tool vibration.An impact damper is a type of particle damper in which masses acts on the tool (Lawrance et al. 2017).The damping mechanisms used in impact and particle dampers are relatively straightforward, cost effective, improved life, and temperature resistant.
A ring with a hole in the centre slightly larger than the diameter of the drill was the form of a damper chosen by Ema and Marui (2000) for their impact damper, which is technically a free mass.The results suggested that this particular damper was successful in reducing tool vibration.Kanase Sandip et al. (2012) proposed a passive damper on the active boring bar to significantly enhance the surface finish of workpiece during the boring operation.Rigid mass was used to dampen and reduce vibration caused by the overhanging of boring bar.Senthil Kumar et al. (2011) studied a mass fixed to the tool holder to reduce vibration in boring tools.Waydande et al. (2014) used mass damping system to reduce vibration and improve surface smoothness during the boring operation.An Impact mass damping system was employed by Sam Paul and Varadarajan (2015) to dampen tool vibration during hard turning.The material is also shown to have a significant impact on the impact mass' damping capacity (Sam and Varadarajan 2014).Using both theoretical and practical approaches, Lawrance et al. (2017) examined the effectiveness of the impact damper on a boring tool.The tool vibrations were effectively reduced using an impact damper, which consisted mass of a predetermined size and form, fixed to the tool holder at certain locations integrated springs.Sathishkumar et al. (2012) have created a particledamping technology that eliminated boring tool chatter and improved surface finish.Filling the boring bar holes with copper also reduced surface roughness by 40%.When applied to a cantilever beam structure, there was a 50% improvement in particle damping (Friend and Kinra 2000).Particle impact damping was successfully extended by Marhadi and Kinra (2005) implementing a variety of materials.To reduce the force of an impact, Diniz et al. (2019) created a particle impact damper, which was stuffed with particles and placed within the boring bar.Vibration was dissipated in the form of frictional energy as a result of particle collusion with one another.The high density material can function effectively as a particle (Khatake and Nitnaware 2013).Impact and particle damping methods have been widely used in structural engineering and metal cutting applications.However, these damping techniques are ineffective at low frequencies.The reason for this is because the low frequency have less energy and this energy is not sufficient to overcome the state function between the materials.As a result, the current study intends to implement BID in the boring bar which was proposed by Li and Darby (2006).To improve the effectiveness when compared with conventional impact and particle dampers, are high impact force and undesirable noise created during the collision of impact and particle mass with rigid ends.But in the case of buffered impact dampers additional soft buffers layer which are flexible were attached to the ends for reducing noise, acceleration and impact force.Thus, a BID allows a high-impact impulse and a suitable exchange of momentum which is to be maintained without the downsides of large impact force (Li and Darby 2009).Real-world issues in academia, the scientific community, commerce, and industry are well-suited to artificial neural networks.Artificial neural networks (ANNs) is a type of logical programming technique that can learn, remember, decide, and infer almost equivalent to the human brain.The artificial neural network architecture is made up of parallel adaptive processing modules which are linked together in a hierarchical-ordered networks.Using ANN, Rao et al. (2016) predicted the RMS velocity and surface roughness of workpiece vibration.Drilling the required diameter of Selective Laser Melted (SLM) Ti6Al4V alloy was studied by Dedeakayoullar et al. (2022), who studied the modelling of surface roughness (Ra) with ANN.Surface roughness optimization during the final turning of AZ61 magnesium alloy was experimentally investigated by Abbas et al. (2018).Badiger et al. (2019) created an artificial neural network model for bushing the length and also adjusted the process parameters through simulated annealing in thermal drilling experiment on galvanized steel.The prediction from the results of an artificial neural network (ANN) showed a good similarity to the experimental results, proving the ANN's to be accurate (Suresh et al. 2021;Kalidass andPalanisamy 2014). Mishra Rohit andBhagat Singh (2023) worked on Spline-Based Local Mean Decomposition (SBLMD) to decompose the sound signals into PFs and reconstruct the new chatter signal.For predicting the CI and MRR, a two, three-layer ANN-based prediction models were built.Statistical comparisons were applied to select the best training algorithm.The ideal cut depth, feed speed, and tool speed ranges were obtained.Mishra Rohit, and Bhagat Singh (2022a) formulated a unique methodology for determining milling stability regimes based on Spline-Based Local Mean Decomposition (SB-LMD) and Artificial Neural Network (ANN).The results from the proposed methodology is appropriate for determining stable milling settings, which would result in improved productivity and a better surface finish.TANSIG with optimum neurons in the hidden layer was identified to be accurate for predicting Chatter Index (CI).Mishra Rohit, and Bhagat Singh (2022b) empirically obtained acoustic signals, three artificial neural network training algorithms, and two activation functions, this research provides a robust methodology for suggesting a stable boundary for greater MRRs in milling processes.The models were optimized using multi-objective particle swarm optimization to find the most suitable range of input values.During the machining process, a modified LMD based on cubic spline interpolation is employed to evaluate the recorded audio signals and to determine the chatter severity in terms of a dimensionless chatter indicator (CI).RSM and MOPSO were implemented to create CI and MRR regression models.The confirmatory experimental results demonstrated that the established optimal solution is viable and can be implemented to increase productivity with chatter-free milling by Mishra Rohit, and Bhagat Singh (2022c).
Artificial neural networks are predominantly studied in various fields of application to predict the behaviour of certain processes and designs which can help the researchers to understand the mechanism of certain characteristics or specific processes.Only few research works were reported on artificial neural networks to predict the tool vibration and cutting parameters in BID assisted boring process.The objective of this study is to predict cutting performance during a buffered impact damper-assisted boring process using an Artificial Neural Network (ANN).There are currently gaps in the literature on predicting cutting performance during the process of using buffered impact dampers, which this study aims to address.The study involves developing an ANN model that can accurately predict cutting performance, including surface roughness, tool wear, tool vibration, and cutting force.The results obtained from the ANN model will be compared with experimental results to evaluate the accuracy and effectiveness of the model in predicting cutting performance during buffered impact damper-assisted boring.

Comparison of tools with and without a damper from a theoretical perspective
The workpiece was assumed to be homogenous and the dynamic parameters of the cutting tool influenced values such as tool vibration and cutting force.

Simulation of the boring process
Mathematical representation of a boring tool holder with a damper, designed for use with a rigid workpiece is represented in Fig. 1.
The cutting force is F(t) and x(t) is the chip thickness (change in depth of cut) at time t.The corresponding mass, damping, and stiffness of the cutting tool and tool holder (m 1 , c 1 , and k 1 ) respectively, is a constant cutting angle, and t represents time.
where, C is the cutting coefficient which is determined experimentally and b is the depth of cut or width of the chip.The formula for the instantaneous chip thickness h(t) is as follows: The chip thickness from the previous machining is denoted by x(t − τ ), the nominal chip thickness produced by the feed mechanism is denoted by h 0 , and the regenerative chatter is represented by the term x(t) − x(t − τ ).Where, the time delay 'τ ' indicates the time interval between each successive passage of the tool, which in boring is equal to the time taken by the work piece to complete one revolution.To obtain the dynamic equation in the Laplacian (s) domain, we equate the Eqs. 2 and 3 into Eq. 1 and apply Laplace transforms on both sides (Juneja and Sekhon 2003).
along with the global transfer function evolving into where

Stability analysis
The spindle speeds (N) and width of the chip (b) are analyzed for studying their correlation using the system's characteristic equation.The characteristic equation for the workpiece analysis is obtained by equating the denominator of Eq. 5 to zero (Juneja and Sekhon 2003).
The following crucial values were derived by solving the values for 'τ ' and 'b' which was obtained by separating the real and imaginary terms when plugging in with s jω.
where n 0,1,2,3…………… The symbol 'ω' denotes the chatter frequency.The formula for calculating spindle speed in revolutions per second is N 1/τ *.Different values of 'n' reveal that Eq. 8 has numerous solutions.Equations 8 and 9 define the boundaries for the stability of the system (Juneja and Sekhon 2003).

Cutting force analysis
Dynamic cutting models are implemented for the analysis of cutting force; at first, the tool deflections are not considered, but as the analysis progresses, the deflection caused by the tool force is considered.Change in deflection of the models with and without damper was determined by applying the results of this analysis.The challenge was formulated using a framework embraced by Juneja and Sekhon (2003).The governing equation for boring when the chip overlapping factor (μ) is set to 1.
Every type of machining results in deformations in one of three shear zones: the primary, secondary, and tertiary.The resultant force can be concluded as F t (tangential forces), F c (cutting forces), and F f (feed forces).

Primary zone
To illustrate the primary zone forces, we have After making the necessary substitutions, the updated equation are in which F s shear force, ∅ c shear angle, α r rake angle, and β a friction angle (Juneja and Sekhon 2003).
Similarly, the expression for normal force is Stress distributed is where, h uncut chip thickness, b cutting breadth

Secondary zone
The rake face of the tool experiences normal force (F v ) and friction force, both of which contribute to the cutting process (F u ) (Juneja and Sekhon 2003).
The coefficient of friction is denoted by From the above equations, we obtain the values for F t and F f , which are expressed below.
The expression for the resultant force is Maximum amplitude solution is obtained by plugging F value into Eq.( 1) and is presented as.
The Damping is accounted for in the above equation, while the without damping equation can be written as

Without damper
The theoretical forces calculation (Table 1), ( 26) Now displacement is given by x max 3.7 mm.

With BID
The theoretical forces calculation, ( 26) x max 0.1228 N .

Material and methods
The commercially available boring tool S25T PCLNR 12F3 is 25mm in diameter and 300mm in length.The workpiece material is AISI 4340 steel hardened to 45 HRC has an outer diameter of F80 mm, an inner diameter of F40 mm, and a length of 100 mm (Kumar et al. 2019).A boring bar equipped with a BID as represented in Fig. 2, the experimental setup, including the input and output parameters, is shown in Fig. 3.The carbon steel boring tool with dimension of F10 mm in diameter and 200 mm in length was machined using spark machining.Particles were loaded into the drilling cavity and fastened in place with an adjustable screw.
The experiment was carried out by 27 test run cycle on a Kirloskar lathe.The parameter such as vibration was measured by piezoelectric vibrometer, the cutting force was measured by Kistler dynamometer, average surface roughness was measured with Mitutoyo-SJ 210, and the tool wear was measured by Toolmakers microscope.For improving the performance, a set of parameters for a buffered impact damper were determined, including the percentage of particle filling in the boring cavity (50, 75, and 100) in which the particle diameter was F4 mm, F6 mm, and F8 mm.The particle

Analysis of variance
The machinability the boring process was studied using Analysis of Variance (ANOVA) to determine the dominance variables and its significance.For determining the influence of each cutting parameter during machining the materials manufacturability, analysis of variance was frequently implemented (Yang and Tarng 1998).Using the software MINITAB 16, the total sum of squared deviations to rank the cutting parameters was estimated with the Eq. ( 27).

Results and discussion
The results from the research work suggested that, buffered impact damper (BID) reduced the vibration, surface roughness, cutting force, and tool wear siginficantally.To mitigate the destructive effects of a collision between a rigid impact mass and a rigid stop while still preserving a control effect, a BID is linked at the ends of the rigid mass and a mass in an effort to dissipate some of the impact force, acceleration, and noise.The installation of the buffer reduces the impact, leading to a considerable decrease in the impact force and an increase in contact time.Further, the buffer material probably has high intrinsic damping, which improves the overall energy dissipation capacity of the structure itself (Li and Darby 2008).The results of the measurements from the study of cutting parameters are represented in Table 3, taken throughout each run.The measured values of the tool vibration and cutting force when using the BID are plotted in Figs. 4 and 5. Surface roughness and tool wear on the BID   and 7.The stainless steel particle content having 75% with minimal particle diameter on a buffered impact damper, provided better cutting performance.
The parameters that contributed for suppression of tool vibration were particle material, particle diameter, and particle filling in boring cavity characteristics and its influence was 37.52%, 32.67%, and 23.88%, respectively.The factors values of data affecting tool vibration and the analysis of variance are shown in Table 4.The particle material which contributes significantly when compared to particle dimension and filling, in lowering tool vibration as per Table 4.Where as from Table 5, on studying the cutting force the particle material (35.27%), particle diameter (33.43%), and filling of particle characteristics (25.07%) all had significant impacts on the cutting force.The particle material contributes 43.32%, the particle dimension contributes 27.99%, and the filling of particle characteristics contributes 22.99% to surface roughness which is evident from Table 6.Table 7 shows the particle material that contributed 38.55% to tool wear, particle diameter contributed 36.7%, and particle filling contributed 20.2%.Considerably very low error percentages were seen for predicting tool vibration (5.91%,), cutting force (6.21%), surface roughness (5.68%), and tool wear (4.45%), respectively.The particle material was found to be the most dominant variable in terms of cutting efficiency and dampening capacity by the ANOVA investigation.The ANOVA analysis was performed with the help of MINITAB16.The effectiveness of damping was found to have affected considerably by the factors such as particle material, particle diameter, and particle filling, as determined by an analysis of variance.
On the basis of experimental data, the input parameters for minimizing the effects of cutting parameter are shown in Table 8.It was found that the particle material of stainless steel, with diameter of 4 mm, and filling of 75% minimized tool vibration and enhanced cutting efficiency.Input parameters of stainless steel particle material are 4 mm particle diameter, and with 75% of particle fills were used in confirmatory experiment against a conventional boring tool.The results from confirmatory experimental are plotted in Table 9.From the results, it is evident that buffered impact damper effectively reduces vibration by 93%, tool wear by 92%, surface polish by 95%, and cutting force by 86%.

Artificial neural network (ANN)
For the present study, the data set for artificial neural networks is derived by conducting 27 experimental runs with three different factors and four responses, as displayed in Table 3 in which the Number of input layers is the particle material, size and filling.There is a range of 2 hidden layers with 50 connectors for the artificial neural network and the output layers are defined as tool vibration, tool wear, cutting force and surface roughness.Algorithm implemented for ANN is Levenberg-Marquardt backpropagation for simulating the network.
The results of an experiment that were randomly divided into three groups and it is displayed in Table 3.About 70% of the data was used for training the network, 15% was used for validation to prevent overfitting the generalized network, and the remaining 15% was used to test the network itself.The architecture of an artificial neural network used to anticipate responses is depicted in Fig. 8. Neurons in each layer are assigned based on the transfer function, which can be linear or nonlinear.In this study for the hidden layers, we use the tan sigmoid function, while the identity function is used for the output layers.
Using momentum backpropagation method, with a constant momentum of 0.98, it was limited to a local minimum.In hidden, the tansig transfer function was used.Both the input and output layers were set from zero-to-one normalized layers.A value of 0.001 was found for both the MSE(mean squared error) and the LRMSE(learning rate mean squared error).After training, the learning rate of neural network was the point at which the MSE was the lowest, and the corresponding network model was then stored.
In the present study, the number of neurons are maintained as 10 and we have optimized 2 hidden layers with 50 connectors for the ANN, and the most acceptable optimum outcome was obtained for the four responses, the 17th, 6th, 13th, and 5th hidden layer for tool vibration, cutting force, surface roughness and tool wear, respectively (Panchal et al. 2011;Stathakis 2009).
We include a number of performance indicators to compare the ANN's projected values for tool vibration, cutting force, surface roughness, and tool wear with the experimental values for these factors.The lists of values for the correlation coefficient (R), mean square error (MSE), absolute average deviation (AAD), and mean absolute percentage error (MAPE) are presented in Table 10.

MSE
1 To achieve the accurate possible outcome, the network is preserved until it attain the best possible hidden layer.The graph plotted in Fig. 9 illustrates the appropriate number of hidden layers, along with the root mean square error (RSME), validation performance (Val performance), number of epochs, and best epoch.It is desirable for the R 2 value to be closer to one, as this would indicate whether the ANN's prediction was accurate.These models have a number of limitations, such as obtaining an accurate numerical expression.Nevertheless, with the advancements in computer facilities, it has become abundantly evident that ANN is the method that delivers the most promising results for creating accurate predictions regarding a variety of factors (Fig. 10).
An ANN makes an inaccurate prediction as a result of overfitting or under fitting the data; however, this problem can be solved through the process of data validation.The untrained data given to the optimal ANN do not perform as the data are over fitted but, similar is not in the case of trained data.When the optimal ANN is unable to predict the trained data, the data are underfitted.The elimination of underfitting in data can be accomplished by adding more   10.
There is a strong influence between the experimental data with ANN predictions and the regression plot for tool vibration, cutting force, surface roughness, and tool wear.
The quality of the best ANN model for various output factors, such as tool vibration, cutting force, surface roughness, and tool wear can be accomplished with the help of the regression plot which is displayed in Fig. 11.It should be noted that the ANN model which was proposed makes accurate predictions regarding the tool vibration, cutting force, surface roughness, and tool wear.

Conclusions
An examination about the use of a BID for the management of tool vibration, tool wear, surface roughness and cutting force was conducted through experimental study.The difference in performance between a buffered impact damper and without damper was analyzed.The following inferences can be drawn based on the results of the tests: • The interactions between the particles and the suspension wall are essential for a buffered impact damper; the stronger the damping, the smaller the particle diameter for improved performance.• Materials with a higher density are preferable because they produce particles in larger impact.On the other hand, hardness was incredibly important too, and it was noticeably evident that hard particles provided superior damping capabilities and cutting parameter values.• The damping equipped by fillings made of 75% packed particles was superior when compared to that equipped by fillings made of 50% or fully packed particles.• Based on the findings of the experiments, it was clear that particles are effective when used for damping capacity.As a consequence of this, making use of a BID is a straight forward and low-cost strategy for lowering tool vibration and other parameters.• From the confirmatory experiments results, the tool vibration was reduced by 93%, cutting force was reduced by 86%, surface finish was enhanced by 95% and tool wear was reduced by 92% for the tool equipped with a buffered impact damper.This research investigates the Artificial Neural Network (ANN) model, which has the potential to aid engineers and other researchers in predicting tool vibration, tool wear, surface roughness and cutting force more accurately.In addition, future research will be focused on additional experimental input parameters as a means of controlling tool vibration within the boring process by making use of the most effective damping mechanisms.

Fig. 1
Fig. 1 A damper-equipped boring tool model for use with a rigid workpiece

Fig. 2
Fig. 2 Buffered impact damper in a boring bar

Fig. 3
Fig. 3 Experimental setup of Buffered impact damper

Fig. 4
Fig. 4 BID on tool vibration during the boring process Fig. 5 BID on cutting force during the boring process

Fig. 8 a
Fig. 8 a Architecture of ANN, b optimal design for ANN

Fig. 9 a
Fig. 9 a Regression plot.b Error histogram plot.c Performance plot.d Train state plot

Fig. 10
Fig. 10 Comparison of experimental data with ANN predicted a Tool b Cutting force, c Surface roughness, d Tool wear

Fig. 11
Fig. 11 Regression plot of experimental with ANN predicted.a Tool vibration.b Cutting force.c Surface roughness.d Tool wear

Table 1
Performance compared with and without the BID damper

Table 2
Representation of parameters of BID

Table 9
Evaluation of the performance of the BID