Experimental study of the effect of different 3D printing parameters on tensile strength, using artificial neural network

Numerous research studies have been conducted to optimize printing parameters using the fused deposition modeling technique (FDM) to improve mechanical properties. The large number of process parameters creates a need to search for optimal combinations of parameters to improve mechanical properties. This study examines the effects of three parameters when printing 3D with melted filament of a PLA material (Polylactic Acid) on the ultimate tensile strength of the printed parts. This search combines an experimental study of the most influential printing parameters on the tensile strength property, namely layer thickness, printing temperature, and feed rate. The experimental results are then analyzed and modeled as a linear regression model. Then develop an intelligent artificial model based on ANN (Artificial Neural Network) derived from these experimental results capable of predicting the optimal combination of parameters providing maximum tensile strength. The observed results showed that the feed rate dominates among the other variables, followed by the thickness of the layer. Also, at the level of prediction, the artificial model provides a better prediction of the tensile strength with a value of 36.1625 MPa by combining the following parameters: Feed rate: 70 mm s−1, temperature: 200 °C, and layer thickness: 0.26 mm, compared to the prediction obtained by the linear regression model. Neural networks enable more accurate optimization of 3D process parameters, leading to an overall improvement in the quality of finished products. predictive models, significantly reducing the iteration time required to obtain optimal parameters. The quality of the data used to train neural networks is crucial.


Introduction
The manufacturing technology known as additive manufacturing (AM) was developed in 1980 for cost-effective and useful use by businesses.This procedure is characterized as forming a portion by adding material [1], rather than typical shaping by removing material.
Additive manufacturing is a generic term that brings together various methods [2], such as fusion deposition modeling (FDM) using plastic filaments [3,4], stereolithography (STL) with a photopolymer liquid [5,6], selective laser synthesis of plastic or metal powders [7,8], and the manufacture of objects laminated by plastic lamination [9].These techniques enable the creation of three-dimensional objects with a variety of materials and offer advantages such as precision, design flexibility, and prototyping speed, thereby revolutionizing industrial manufacturing processes.
The fusion deposition modeling technique involves the creation of objects by adding successive layers of thermoplastic melted material, usually in the form of a filament heated to a temperature where it becomes liquid.The melting material is extruded through a mobile hose on a construction plate.The hose moves according to a predefined pattern to lay down the material in successive layers.These printers are widely used and have gained great importance in the additive manufacturing market due to their advantages [10].The advantage of this technology over traditional procedures is its ability to produce complicated pieces [11,12] in less time and at a lower cost [13].Unfortunately, this approach still has drawbacks in terms of end output, which motivates researchers to enhance them through printing condition optimization [14].
The mechanical properties of parts printed using additive manufacturing techniques continue to be a need sought by authors and research laboratories [15]; the parameters that influence these properties are still poorly understood due to a large number of parameters such as infill density Print temperature, orientations, print speed, bed temperature, infill pattern, shell thickness, nozzle diameter, and filament, as well as the time and cost of each prototyping experiment, and experimental protocol for each property.
In previous experiments, for example, the study [15], the authors examined the influence of print orientation and filling density on the tensile strength of 3D made parts.The study confirmed that the optimal print orientation is on-edge.On the other hand, the study [16] tested the effect of infill pattern and layer thickness on tensile strength.The results of this study show that mechanical properties are more affected by the thickness of the layer than by the filling pattern.The thickness of the layer increases mechanical strength, and the triangular pattern gives the highest strength.
The current study aims to integrate artificial intelligence tools to predict and optimize these properties, as well as to find the link between these properties with the greatest number of parameters.
The literature review in this study shows that printing parameters in fused deposition modeling (FDM), such as layer thickness, printing temperature, and feed rate, are the most influential print parameters on the tensile strength property.This study combines these parameters to study the importance of those parameters by printing a series of test samples that represent different combinations.The results of the experiment are then analyzed and modeled using a linear regression model.Then, from these experimental results, create an intelligent artificial model based on ANN.
Using an ANN network of artificial neurons to optimize 3D printing parameters [17] is an interesting approach that falls within the field of machine learning.It takes into account complex relationships between variables, which can be difficult to do analytically [18].However, it is important to note that this requires significant data collection and machine learning expertise to develop and adjust the model properly [19].In addition, the use of AI to optimize 3D printing parameters can greatly facilitate the process of adjusting the parameters to get the best performance.

Review of literature
To comprehend and evaluate the authors' articles and previous studies, as well as the influence of each parameter, a preliminary literature search is required.The study that follows provides a summary of the research on the impact of infill density, layer thickness, and printing orientations on the mechanical properties of printed materials.
Twenty five papers  conducting this review of literature were obtained by using 'Science Direct', 'Springer Link', 'Scopus', and 'Sage' search databases.
Journal articles represent 87% of selected references, followed by conference papers with 13%.The papers pulled from these databases are distributed as figure 1 shows, Scopus in first place with 50% of the journal articles, followed by Science Direct in second place with 36% of the references, Springer Link, and Sage in last place with 9% and 5%.
The analyzed references are from the years 2017-2023, as shown in figure 2. The bar graph demonstrates that the highest frequency is in 2020.The year 2023 ranked second with six mentions of references.As illustrated in figure 3, extracted from the keyword analysis of this database, the additive manufacturing keyword is highlighted, along with the ANN model, 3D printing parameters, infill density, shell thickness, PLA, and process.These words underline the central purpose of this literature review, highlighting key elements and specific areas related to 3D printing and additive manufacturing.
Thematic database analysis involves classifying 3D printing parameters according to their respective indices.This methodology allows a structured organization of the data, facilitating the understanding of the relationships between the different parameters and their implications in the context of 3D printing, table 1 shows the frequency of evaluation of each parameter in all articles covered in this review of literature.For example, infill density has been treated in fifteen articles out of twenty-five, layer thickness has been treated in thirteen articles, and print temperature has been processed in ten articles.Each parameter has an indices that represents the range of value of this parameter; for example, for infill density, the fifteen articles treated examine that parameter within a range of 10 to 90%.The most significant 3D printing parameters shown in figure 4 are based on FDM technology and 3D printable material.For infill density, this is the most processed parameter fifteen times, which represents 20% of the total parameter processing frequencies, followed by layer thickness with 18% and print temperature with a percentage of 14%.

Material and methods
In this study, each sample was printed with a different combination of parameters, with the main purpose of analyzing the influence of layer thickness, printing temperature, and feed rate parameters on the tensile strength as well as on the production time.
The experiment takes place according to the flowchart in figure 5.The objective of the bibliographic study is to investigate the tensile strength properties of 3D printed parts.The results of these articles give an overview of the importance of each parameter.It is noteworthy that the combination of parameters: layer thickness, printing temperature, and feed rate has not been evaluated, so the choice to do this study with these parameters.
In the first step, it is necessary to build a 3D model with CAD software (CATIA V5 software), according to the dimensions [45] represented in figure 6.The printing is done with a 3D printer, as depicted in figure 7.Each sample has been printed with a combination of different parameters.After the samples were printed, a tensile

Equipment and process parameters
The fusion deposition modeling technique is a thermal procedure that uses a heating system to melt a PLA filament, which is then deposited on the moving printing table by an extrusion nozzle.
PLA is a biodegradable, the significance of studying PLA material lies in its enhanced mechanical and thermal properties.As the most widely utilized material in the additive manufacturing of plastic materials, PLA holds paramount importance.Table 2 represents the mechanical properties of PLA [45].
The samples were printed using a 3D printer in line with ASTM D638, using various combinations of layer thickness, printing temperature, and feed rate while maintaining constant infill density, printing bed temperature, nozzle diameter, print orientations, and infill pattern.
Layer thicknesses of 0.16 mm and 0.28 mm were selected.The printing temperatures used to print are: 200 °C, 215 °C, and 230 °C; the chosen feed rates are 40 mm s −1 , 65 mm s −1 , and 70 mm s −1 ; the chosen printing orientation is on-edge, giving the best strength result [34,44]; the chosen infill density parameter is 100%; this parameter is very important because it affects the mechanical strength of the sample, which increases with the chosen maximum infill density [31,39].The temperature of the heating bed was adjusted to 60 °C, the diameter of the nozzle is 0.4 mm, the infill pattern is line concentric [42], and the summary of process parameters is shown in table 3.

Design and 3D printing
Dog bones-haped tensile sample were designed in the CATIA program and saved in STL format.After that, the STL files were transferred into the Ultimaker CURA software [46], which sliced and transformed them into GCODE format.
Figure 6 depicts the dimensions of the standard specimen [45] for mechanical testing.(All measurements are in millimeters).
Many 3D printing technologies are now on the market, and many more are in the works.The 3D printer available in the lab is a professional printer, Ender-3 figure 7, with a feed rate of up to 180 mm s −1 , an operating mode of online or offline with a Secure Digital (SD) card, a nozzle temperature of up to 255 °C, a nozzle diameter of 0.4 mm, and a layer thickness available between 0.1 and 0.4 mm.
The following table 4 summarizes the combination of the parameters considered: In order to give originality to this study, it is necessary to find new combinations of parameters that can give new results while not exceeding the margins of the most commonly used parameters for validation and comparison.This is why, following the review of literature, it turns out that this selection of parameters from table 4 meets these criteria.

Experimental approach and mechanical testing of samples
The test machine available in the laboratory is a professional tensile machine 3 R Syntax, figure 8, with a capacity of 100 kN.The machine has an interface connected to a computer, and the software 3 R Syntax allows the tracing of the curve of the tensile stress as a function of deformation and the generation of Excel files for each test containing all the information concerning the test, such as the time of the test, the force, and the measurement of the extensile stress.All tests were carried out in the same ambient conditions at a temperature of 22 °C and 50% humidity at a test speed of 2 mm min −1 .Before carrying out the mechanical tests, the specimens were manually numbered at both ends.Figure 8 shows the specimens being held between the machine's jaws.Then, applying a slight pre-load of 100 N removes any matching or sliding in the sample before recording the data during the test, thus ensuring accurate and reproducible results.During the tensile test, the 3R Syntax software evaluates the deformations in relation to the force delivered.The graphs are then represented, and the tensile strength is calculated using the following definition, equation (1).
Stress at fractured σr (MPa): Figure 9 illustrates the 18 fractured specimens following the test; the 18 tests were performed one after the other in the same room under the same conditions to avoid any influence of external variables, temperature and humidity, on the predicted findings.

Results of experimental work
The numerical results of the tensile tests on the specimens are listed in table 5, and for each combination of the input parameters examined, the printing time and weight of each specimen have been incorporated.
Production times for samples with high feed rate have decreased significantly, as shown in table 5, which means that manufacturing costs will decrease as the feed rate increases.
The effects of the feed rate on the tensile strength of the first six samples printed at a speed of 40 mm s −1 are illustrated in figure 10.In this figure, it is noted that with the increase in layer thickness and the decrease in printing temperature, the value of tensile strength increases.Sample no. 2 with a layer thickness of 0.28 mm and a printing temperature of 200 °C has the maximum value of tensile strength, followed by sample no.1 with a layer thickness of 0.16 mm and a printing temperature of 200 °C.
For the second series of samples from 7 to 12, which are printed at a feed rate of 55 mm s −1 , the results are shown in figure 11.Sample no.12 with a layer thickness of 0.28 mm and a printing temperature of 230 °C has the maximum value of tensile strength, whereas sample no.7 with a layer thickening of 0.16 mm and printing temperatures of 200 °C shows the low tensile strength in this series.Finally, figure 12 shows the tensile strength results for samples from 13 to 18; it is noted from the diagram that sample no.16 presents the maximum tensile strength in this series with layer thickness parameters of 0.28 mm and a printing temperature of 215 °C, while sample no. 13 presents the low tensile strength value.
It is also observed in all tests that increasing the feed rate causes an increase in the value of the tensile strength.The highest values of tensile strength are given by the 70 mm s −1 sample series.
In all samples, the effect of the change in printing temperature is very small.Increasing the temperature to 230 °C has less effect on tensile strength than 200 °C, which means that energy consumption in heating resistors can be reduced by minimizing the extrusion temperature.
Finally, it is also noted that with the increase in layer thickness, the tensile strength increases in a remarkable way, which increases the material deposit flow rate, which directly affects the production time, which will be reduced, and which will also minimize manufacturing costs.
An analysis of these results was carried out using the Minitab software; figure 13 created by Minitab represents the graph of the main effects, allowing us to examine the differences between the averages of the levels of these factors : feed rate, layer thickness and print temperature, giving the direct impact of each parameter on the evaluated output.
In figure 13, it is noteworthy that increasing the feed rate from 40 mm s −1 to 70 mm s −1 leads to an improvement in the tensile strength of the sample.Similarly, increasing the layer thickness also implies an  increase in tensile strength.With regard to the printing temperature, the effect of this parameter is less significant.The increase in temperature from 200 °C to 230 °C did not result in much change in the value of tensile strength.
A regression analysis was performed using the Minitab software to establish a mathematical regression model.Regression analysis generates an equation to describe the statistical relationship between one or more predictors and the response variable, as well as to predict new values of the output variable.This linear regression Minitab uses the least squares estimate method, which determines the equation by reducing the sum of squared residual values.Equation (2) presents the regression model that gives the behavior of the tensile strength (Ts) based on the evaluated parameters: feed rate (Fr), temperature (T), and layer thickness (Lt).This equation is made up of products and sums of constants such as 26.82; 0.0846; 0.0027 and 9.76.

= +
´-´+ ´( ) Ts 26.82 0.0846 Fr 0.0027 T 9.76 Lt 2 A comparison was made in figure 14 between the tensile strength test results and the results calculated using established regression models.The graph allows you to compare the trends between the two curves that follow the same trend in this case.The graph also allows you to visually control the adjustment of the weighted averages to the data through smooth curves.This type of average is generally used as a method of smoothing values, removing transitory fluctuations, in order to highlight trends.
The chart shows a decrease in tensile strength, more specifically in test no.6.The reason is that, in general, 3D-printed parts present an inhomogeneity of printed material, which can decrease the strength of the sample.This decrease is reduced by the regression model.

Artificial neural network modelling
Following their improved prediction capacity [47,48], the artificial models realized by neural networks are applied for the prediction of the output, tensile strength, based on the results of the experimental tests.Table 5, the constructed model takes as inputs the layer's thickness, the printing temperature, and the feed rate.
MATLAB software has been used for the construction, training, testing, and validation of the ANN model.According to documents [17,26] and [48] the values of the data reports have been retained.70% of the data is for training, 15% for testing, and 15% for validation.The diagram is constructed with two layers, a hidden layer of 20 neurons that receives input data with a Logsig activation function, and an output layer with a Purelin activation function.
The diagram of the constructed model is represented in figure 15, and the choice of the parameters of this network is justified by the article [49,50].These parameters are:  Figure17 shows the evolution of the mean square error of the ANN model during training.When analyzing this evolution, it is observed that the mean square error decreases with the iterations performed during training.When the validation mean square error starts to increase, iterations stop automatically.The best validation performance achieved is 1.0308e −6 .
In figure 18, a comparison between the two models already established, the regression model, and the ANN model, in relation to the tensile strength measured by the test, the analysis of figure 18, a remarkably better forecast is observed by the ANN model.
Analysis of this figure leads to the added value of using an artificial neural network model (ANN), which lies in its powerful ability at the learning level and its ability to capture complex, nonlinear relationships between input and output data.Unlike simpler models, ANNs can learn abstract models from data, which makes them effective for prediction and generalization.This study allowed us to investigate the influence of the parameters considered and evaluated in these experiments.Table 3 sets the values of some parameters justified by the literature: the filling density at 100%, the orientation of the on-edge construction, and the concentric linear pattern infill.This study assessed other important parameters, such as printing temperature, layer thickness, and feed rate, which allowed us to give a general estimate of tensile strength with the maximum printing parameters.It can be summarized that the contribution of this study is divided into two parts.The first is to give an optimal combination that combines a series of 3D printing parameters.The data established in these experiments can help in the future objective of investigating the influence of all parameters on the tensile strength of 3D-printed PLAs.The second contribution is the use of an ANN prediction model to optimize tensile strength, which introduces all these experimental data.

Conclusion
The results of the work presented in this study present another rich experimental database that adds to the process of research on the evaluation of the mechanical properties of 3D printed parts.The data and the results established could be used in the future for improvements and critical innovations in the quality of printed prototypes.
This experimental contribution indicates an assessment of tensile strength following a series of combinations of parameters.The first observation is that tensile strength increases with increased feed rate, while the change in printing temperature does not have a great effect on the tensile strength.An increased feed rate has a beneficial effect on production time and cost, thus minimizing it.Thus, the effect of print temperature also minimizes energy consumption in relation to production.
The use of prediction with the artificial neuron network model (ANN) allows to choose an optimal combination of process parameters and to conclude on the choice of layer thickness parameter, which should be higher.The developed artificial ANN model allows for the prediction of tensile strength for any combinations of these input parameters and minimizes the number of destructive experiments.The performance of this ANN model quantified by a Mse = 1.0308e −6 and by the R = 1 correlation coefficients of training, validation, and testing guarantees the reliability of this model in the prediction and its ability to generalize it to new values of the input parameters.
The expected perspective is to build a database of experimental data that deals with the influence of all parameters of the 3D printing process on tensile strength and to generalize this ANN prediction model for maximum input parameters, giving a powerful model that can predict tensile strength without the need for destructive experience.

Figure 1 .
Figure 1.The distribution of databases.

Figure 2 .
Figure 2. Period and frequency of publication.

Figure 4 .
Figure 4.The more influencing parameter of the 3D printing technology.

Figure 5 .
Figure 5.The flowchart of the methodology.

Figure 10 .
Figure 10.Tensile strength curves for specimens with a feed rate of 40 mm s −1 .

Figure 11 .
Figure 11.Tensile strength curves for specimens with a feed rate of 55 mm s −1 .

Figure 12 .
Figure 12.Tensile strength curves for specimens with a feed rate of 70 mm s −1 .

Figure 14 .
Figure 14.Tensile strength and Tensile strength by regression model.
After creating, training, and validating an artificial neural network model (ANN), the ANN model is ready to receive new random combinations of input parameters.It is possible to predict tensile strength for new combinations of parameters from a series of 448 random combinations.Figure19 shows the answers calculated by the ANN model, with a predicted maximum value of 36.1625MPa corresponding to the following combination no.430: feed speed: 70 mm s −1 , temperature: 220 °C, layer thickness: 0.26 mm.

Figure 18 .
Figure 18.Tensile strength, Tensile strength by regression model and Tensile strength by neural network (ANN).

Figure 19 .
Figure 19.The predicted tensile strength for the new combinations.

Table 2 .
Conditions for 3D printing of PLA.

Table 5 .
Results of mechanical and printing tests (for Samples 1 through 18).