AI-Based Surface Roughness Prediction for Additively Manufactured Components

Version	Summary	Created by	Modification	Content Size	Created at	Operation
1		Temesgen Batu Gemechu	--	3916	2023-10-21 08:17:20	\|
2	Reference format revised.	Lindsay Dong	Meta information modification	3916	2023-10-23 05:10:25	\|

This entry is adapted from the peer-reviewed paper 10.3390/ma16186266

Additive manufacturing has gained significant popularity from a manufacturing perspective due to its potential for improving production efficiency. However, ensuring consistent product quality within predetermined equipment, cost, and time constraints remains a persistent challenge. Surface roughness, a crucial quality parameter, presents difficulties in meeting the required standards, posing significant challenges in industries such as automotive, aerospace, medical devices, energy, optics, and electronics manufacturing, where surface quality directly impacts performance and functionality. Artificial intelligence (AI) is one of the methods used by researchers to predict the surface quality of additively fabricated parts.

surface roughness roughness prediction additive manufacturing machine learning

1. Introduction

Over recent decades, the manufacturing sector has gone through significant changes, yet it continues to play a critical role in the growth and development of both developed and developing countries ^[1]. It is considered the backbone of economies due to its contribution to economic growth, job creation, and technological advancement ^[2]. Additive manufacturing (AM) has gained acceptance among academics and the manufacturing sector as a powerful manufacturing tool. Recent research shows that it is more efficient than conventional production methods ^[3]. The manufacturing industry uses the term Additive Manufacturing to describe the process of fabricating physical objects from design data in a digital form by building them layer-by-layer. AM is a disruptive technology as it is not tied to individual production steps and does not require specific tools for each component, making it a universal production technology ^[4]^[5]. Numerous AM processes are currently available for commercial use. Various researchers have approached the classification of AM methods differently ^[6]^[7]^[8]. However, a widely accepted classification stems from the ASTM-F42 committee guidelines, which categorize AM into seven distinct groups ^[9]. These categories consist of vat photopolymerization (VP), material jetting (MJ), binder jetting (BJ), material extrusion (ME), sheet lamination (SL), powder bed fusion (PBF), and directed energy deposition (DED). The concise overview of all seven categories was presented in the paper ^[10].

The main advantage of additive manufacturing is that it can fabricate complex shapes and reduces material waste and production time (for some) ^[11]^[12]^[13]^[14]. Despite its advantages, additively manufactured products generally have poorer quality compared to those made with conventional manufacturing systems ^[15]^[16], primarily due to limitations in surface integrity ^[17]. Research indicates that in all additive manufacturing techniques, the improvement of surface roughness is a key objective. For example, vat photopolymerization (VP) and material jetting (MJ) yield parts with moderate surface roughness ^[10], whereas binder jetting (BJ), material extrusion (ME), sheet lamination (SL), powder bed fusion (PBF), and DED tend to result in parts with relatively poorer surface finishes ^[10]. Surface roughness is one of the factors encompassed by the concept of surface integrity.

Surface roughness has been identified as one of the most significant factors affecting product quality, as stated in previous research ^[11]. For instance, Chan et al. ^[18] conducted a study to investigate the impact of surface roughness on product life and concluded that surface roughness leads to a reduction in product life expectancy. Moreover, surface roughness plays a role in the tribological behavior of surfaces ^[19], with rough surfaces experiencing faster wear compared to smooth surfaces. Therefore, it becomes crucial to predict and control the surface roughness of additively manufactured parts ^[20]^[21]. Additionally, surface roughness serves as an indicator for directly monitoring the mechanical characteristics of a workpiece, such as fatigue and surface friction, dimensional accuracy ^[22], and fracture resistance ^[23]. Understanding and managing surface roughness in manufacturing processes is essential for ensuring optimal product performance and longevity.

Different methods were used to predict the surface roughness of additively manufactured components, i.e., Taguchi-based regression models ^[24]^[25], statistical regression models ^[26], computational modeling (e.g., the FEM and Discrete Element Method (DEM)), ^[27] and machine learning methods ^[28]. There are several limitations of conventional (statistical) methods for predicting the surface roughness of additively manufactured components. Conventional methods for predicting the surface roughness of manufactured components often suffer from various limitations, such as a lack of flexibility, time-consuming nature, high cost, limited accuracy, limited applicability, inability to account for process parameter interactions, and limited understanding of underlying mechanisms ^[17]^[29].

2. AI-Based Surface Roughness Prediction

2.1. Artificial Intelligence Overview

Artificial intelligence is defined as a subset of IT applications that can capture information from its environment, comprehend, learn, interpret, and derive actions based on their implemented objectives ^[30]. In its strictest definition, AI stands for the imitation by computers of the intelligence inherent in humans ^[31]. Machine Learning, a prominent subset of AI, falls under the narrow AI category. It is characterized by its task-specific nature and focuses on a limited range of functionalities. Machine Learning algorithms are designed to learn from input data and enhance their performance in specific domains such as image recognition or natural language processing ^[32]. Machine Learning (ML) is a subfield of AI that emphasizes the development of algorithms and models enabling machines to learn from data and make predictions and decisions without explicit programming. In other words, ML involves training algorithms on extensive datasets to identify patterns and relationships, which are then utilized for making predictions or decisions ^[33]. ML’s primary objective is to analyze and learn from given datasets to perform tasks. It is classified into three categories: (1) supervised, (2) unsupervised, and (3) reinforcement learning. Figure 1 provides an overview of common ML approaches.

Figure 1. Taxonomy of machine learning applications in the AM domain. (Adapted from ^[33] an open-access article distributed under the terms of the Creative Commons CC-BY license).

2.2. Surface Roughness Prediction

2.2.1. Definition of Surface Roughness and Its Measurements Techniques

Consist of profile, form, waviness, and roughness components, each with distinct origins and effects on product appearance and functionality. Waviness exposes machine vibrations, form arises from manufacturing system limitations, profile relates to layer-by-layer manufacturing, and roughness stems from printing and material removal errors. Waviness, akin to signal noise, emerges due to motion system planarity and deformations caused by weight or residual stress ^[34]. Process-specific factors, such as defects, thermal distortion, adhesion issues, support structure inadequacies, and post-processing deformation, also contribute to waviness ^[35]^[36].

Surface roughness, a crucial texture element, evaluates manufactured item quality by assessing topographical feature distribution. Diverse metrics are employed across industries, addressing uncertainty in 3D-printed product surface quality through multiple metrics for efficiency ^[37]. For instance, for the area surface roughness evaluation, the average area roughness (S_a) and area root mean squared height (S_q) are insensitive to measurement parameters ^[38]. Area height distribution skewness (S_sk) effectively characterizes surfaces in SLM parts. Surface roughness impacts microstructures with peaks and valleys of varying heights, which is crucial as components miniaturize ^[39].

2.2.2. AI-Based Prediction of Surface Roughness

Figure 2 illustrates the general approach to develop data-driven predictive models that can be used for surface roughness estimation. It illustrates the presence of input, which comprises condition monitoring data, along with predictive modeling, ultimately resulting in the output of surface roughness. These input data, either independently or in combination, are employed to train machine learning models for predicting surface roughness in additively manufactured components. It is crucial to emphasize that both the quality and quantity of the data significantly impact the accuracy and reliability of the predictions. Therefore, the careful collection and processing of data are vital to prevent any potential biases or errors.

Figure 2. Data-driven predictive modeling.

The prediction of surface roughness for additively manufactured components is a fundamental application of machine learning, where data-driven modeling is used to predict the surface roughness of additively manufactured components ^[40]. There are several types of data that can be utilized for this prediction. Commonly used data types include:

Process parameters: These include data related to the additive manufacturing process. In LPBF, for example, parameters such as laser power, scan speed, layer thickness, and hatch spacing ^[41] have been used for predicting surface roughness.

Material properties: These include data related to the material used in the additive manufacturing process, such as powder size, particle shape, and chemical composition, especially in methods such as L-PBF for metal AM. Material properties can also have a significant impact on surface roughness ^[42].

Geometrical features: These include data related to the geometry of the final product, such as the angle and direction of the surface, the size and shape of the features, and the number of layers. Geometrical features can influence the surface roughness by affecting the thermal and mechanical properties of the material. For example, in FDM, building orientation has been employed to predict surface quality ^[43]. In the case of EBM, on the other hand, printed components, the sloping angle and surface orientation have been used to predict surface roughness ^[44].

Environmental conditions: These include data related to the ambient conditions during the additive manufacturing process, such as temperature, humidity, and pressure. These conditions can affect the material properties and, in turn, the surface roughness of the final product ^[21].

Surface texture data: These types of data involve measuring the surface roughness of an additively manufactured component using specialized instruments such as profilometers or surface roughness testers. The surface texture data can be used as a training dataset for machine learning models to predict surface roughness based on process parameters, material properties, geometrical features, and environmental conditions ^[45]^[46].

Vibration data: Vibration data play a crucial role in predicting surface roughness for additively manufactured components. By capturing and analyzing the vibrational characteristics during the additive manufacturing process, valuable insights can be gained regarding the quality and surface roughness of the manufactured components. Vibration data can help identify any anomalies, such as excessive vibrations or oscillations that may affect the surface roughness.

3. AI-Based Surface Roughness Prediction Methods for Additively Manufactured Parts

3.1. Traditional Machine Learning Approaches

It should be emphasized that conventional machine learning techniques offer a strong basis for surface roughness prediction. However, they might face challenges in capturing intricate non-linear connections or managing data with high dimensions. The application of deep learning techniques has gained prominence in recent years due to their ability to handle these challenges more effectively. The common traditional machine learning algorithms that can be applied for the surface roughness prediction of additively manufactured components are Linear Regression, Support Vector Machine, Random Forests, Gradient Boosting Algorithms, and Artificial Neural Networks.

3.1.1. Support Vector Machines

A surface roughness prediction in the context of additive manufacturing can be facilitated through various supervised learning algorithms. One such algorithm is SVM, which aims to find an optimal hyperplane for data classification. SVM is widely used due to its ability to handle both linear and nonlinear data, as well as its high accuracy ^[47].

Singh et al. ^[48] applied SVM techniques to model the Wire Electrical Discharge Machining (WEDM) process of AA6063 for armor applications. They considered four input variables: pulse-on-time (Pon), pulse-off-time (Poff), servo-voltage (V S), and peak-current (IP), with surface roughness as the response parameter. By employing a 3k full factorial design for the experimental runs, the developed model demonstrated its predictive capability and suitability for smart manufacturing. The surfaces of the machined components were further evaluated using SEM analysis.

3.1.2. Random Forests

Random forests are a type of ensemble model that combines multiple decision trees. Each tree is trained on a different subset of data and features, and the final prediction is obtained through a voting or averaging process. The strength of random forests lies in their ability to handle complex relationships and interactions among input parameters, making them well-suited for surface roughness prediction ^[49].

Li et al. ^[17] utilized a random forest (RF) algorithm as part of an ensemble learning-based approach for surface roughness prediction in Fused Filament Fabrication (FFF) processes. The authors incorporated multiple sensors to gather real-time condition monitoring data and extracted a set of features from the raw sensor-based signals in both the time and frequency domains. To enhance computational efficiency and mitigate overfitting, the researchers employed RF to select a subset of 40 features based on their importance. The ensemble learning algorithm in the study combined six different machine learning algorithms, including RF, AdaBoost, CART, SVR, RR, and the RVFL network. The experimental results demonstrated the predictive models’ capability to accurately predict the surface roughness of 3D-printed specimens. The ensemble model outperformed the individual base learners, as evidenced by the lower Root Mean Square Error (RMSE) and Relative Error (RE). The authors concluded that this ensemble learning-based approach, incorporating RF as a feature selection method, shows promise for predicting the surface roughness of additively manufactured components in other processes such as selective laser sintering and electron beam melting.

3.1.3. K-Nearest Neighbors

K-nearest neighbors (KNN) is a non-parametric algorithm that utilizes proximity to the nearest neighbors to classify or predict data points. In the context of surface roughness prediction, KNN can estimate roughness by identifying the nearest neighbors with similar input parameter values ^[50].

Kumar and Jain ^[45] focused on employing the KNN machine learning algorithm for surface roughness prediction. The authors generated training data for the KNN algorithm by depositing multi-layer single-track depositions, resulting in wall-like structures, using Stellite-6 as the additive manufacturing material in both powder and wire forms. Their findings revealed that surface roughness increases with a higher power supply to the micro-plasma and AM material feed rate, while it decreases with an increase in the traverse speed of the deposition head for both powder and wire forms of the AM material. Additionally, the surface roughness of the walls produced with the powder form of the AM material (ranging from 118 to 149 μm) was found to be smaller than that obtained with the wire form (ranging from 195 to 227 μm). The prediction error of surface roughness using the KNN algorithm ranged from −6.2% to 2.8% for the powder form and −5.8% to 2.3% for the wire form of the AM material. These results demonstrate the capability of the KNN algorithm in accurately predicting surface roughness in the μ-PTAMAM process. Furthermore, the authors highlighted that increasing the number of training datasets can further reduce the prediction error of the KNN algorithm.

3.1.4. Artificial Neural Networks (ANN)

ANNs are fascinating machine learning algorithms inspired by the intricate structure and functionality of the human brain. Composed of interconnected nodes called “neurons”, ANNs possess the remarkable ability to process and transmit information. These networks find applications in a wide range of tasks, including pattern recognition, image and speech recognition, prediction, and classification ^[51].

One captivating application of ANNs lies in the prediction of surface roughness for additively manufactured components ^[52]. Additive manufacturing, a process that builds 3D objects layer by layer using computer models, presents a challenge in controlling the surface roughness of printed objects—a crucial factor affecting their mechanical properties and overall performance. To address this challenge, ANNs can be trained on datasets comprising input variables and corresponding surface roughness measurements. These input variables encompass factors such as the type of printing material, layer thickness, printing speed, and printing temperature. As ANNs learn from the data, they identify patterns and make predictions based on new inputs. Once trained, ANNs can accurately predict the surface roughness of new additively manufactured components based on their input parameters. This predictive capability empowers manufacturers to optimize their printing processes and enhance the quality of their printed parts ^[53].

ANN have been widely utilized by researchers to predict the surface roughness of additively manufactured components. In a remarkable study conducted by Wafa and Abdulshahed ^[54], they employed an ANN approach to accurately predict the surface roughness of FDM-printed components. The ANN model, built with a small number of neurons in the MATLAB environment, exhibited exceptional agreement with the experimental data, achieving an average error value of only 8%. Furthermore, the researchers compared their proposed ANN model to a regression-based approach and found that the ANN model outperformed the statistical method in terms of accuracy.

3.1.5. Others Machine Learnings

There are other machine learning algorithms used for surface roughness prediction. For instance, a machine learning method based on Gaussian Process Regression was proposed to establish a model relating the WAAM process parameters to the top surface roughness. To measure the top surface roughness of a manufactured part, a 3D laser measurement system was developed. The experimental datasets were collected and subsequently divided into training and testing datasets. Using the training datasets, a top surface roughness model was constructed and then verified using the testing datasets. The experimental results demonstrate that the proposed method achieves a surface roughness prediction accuracy of less than 50 µm ^[55].

Li et al. ^[17] introduced the ensemble learning algorithm to determine predictive models for surface roughness. Their approach involved training the model using different learning algorithms, including random forests, AdaBoost, classification and regression trees (CART), SVR, RR, and random vector functional link (RVFL) networks. The experimental results demonstrated the effectiveness of this ensemble learning-based approach, particularly in accurately predicting the surface roughness of fused filament fabrication (FFF)-manufactured components.

3.2. Deep Learning Approaches

Machine learning techniques are considered a weak AI type, which means they are not entirely autonomous and require some level of guidance, such as adjusting hyperparameters. Deep learning (DL) methods were developed to push beyond the limitations of traditional machine learning, which are also aimed to emulate certain aspects of human cognition more closely. Consequently, they have proved to have a superior performance in terms of accuracy and speed compared to other machine learning algorithms, without the need for significant manual intervention from programmers. Deep learning is a specific subset of machine learning that operates by processing inputs through a biologically inspired ANN architecture. Over time, it has become evident that neural networks outperform many other algorithms in accuracy and speed due to their powerful ability to extract relevant information from vast amounts of data ^[56].

As can be observed from Figure 3, the main difference between deep learning and traditional machine learning lies in the architecture and representation of data. In traditional machine learning, feature engineering is a separate step where domain experts manually select and engineer relevant features from the raw data. While traditional methods may be more interpretable, they heavily rely on domain knowledge and might struggle to capture intricate relationships or patterns in the data without comprehensive feature engineering. Moreover, the feature engineering process can be time-consuming and subjective. However, in deep learning, feature extraction and prediction are performed in an end-to-end manner using neural networks. Raw data are directly fed into the network, and the model automatically learns hierarchical representations and complex features, eliminating the need for manual feature engineering. The neural network’s layers capture abstract patterns and relationships within the data, enabling it to predict surface roughness directly from the raw input, making the process more efficient and accurate with a large amount of labeled data. However, the black-box nature of deep learning models may limit interpretability ^[57].

Figure 3. Traditional machine learning vs deep learning for surface roughness prediction.

Deep learning possesses remarkable capabilities for modeling and handling highly intricate non-linear relationships. Numerous variants of deep learning techniques are available, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), artificial neural networks (ANNs), and deep neural networks (DNNs). These neural networks consist of artificial neurons organized in multiple layers, where each layer communicates only with the immediately preceding and following layers (as shown in Figure 4).

Figure 4. Schematic representation of a deep neural network, a type of artificial neural network featuring several hidden layers of neurons between the input and output layers.

3.2.1. Convolutional Neural Networks

Convolutional Neural Networks (CNNs) are widely used in image recognition and classification tasks, particularly in the analysis of surface topography scans and visual representations of additively manufactured components. These networks utilize convolutional layers to extract hierarchical features from input data, allowing them to capture intricate patterns and textures related to surface roughness. Researchers have recently employed CNNs for predicting the surface roughness of additively manufactured components ^[58].

In a recent study ^[59], a deep learning CNN model was utilized to predict the surface roughness of additively manufactured components. The study proposed a combined approach of CNN classification and electrical discharge-assisted post-processing to enhance the surface quality of these components. By categorizing the surface, the depth and number of polishing passes were determined. The study revealed that polishing under a low-energy regime outperformed high-energy regimes, resulting in a significant 74% improvement in surface finish. Additionally, lower-energy polishing reduced the occurrence of short-circuit discharges and elemental migration. The CNN model demonstrated 96% accuracy in predicting the surface condition through a five-fold cross-validation. Furthermore, the proposed approach substantially improved the surface finish from 97.3 to 12.62 μm.

3.2.2. Deep Neural Networks

A deep neural network (DNN) is an ANN having multiple layers between the input and output layers. A DNN consists of neurons, synapses, biases, weights, and functions. Deep neural networks have demonstrated discriminative and representative learning capabilities across various applications in recent years ^[60]^[61].

So et al. ^[11] developed a method to improve the quality of additively manufactured products by predicting surface roughness using data analysis techniques, including data pre-processing and DNNs combined with sensor data. The study focused on enhancing the surface roughness of the stacked wall, which is a crucial quality indicator affecting product life and structural performance. By applying data pre-processing and DNNs with sensor data, the study proposed a methodology to predict surface roughness based on process parameters. The effectiveness of the proposed methodology was validated using field data from wire + arc additive manufacturing, resulting in a mean absolute percentage error (MAPE) of 1.93%.

3.2.3. Generative Adversarial Networks

Generative Adversarial Networks (GANs) have shown promise in predicting the surface roughness of additively manufactured components by generating synthetic profiles that closely resemble real-world data. GANs consist of a generator network and a discriminator network that compete against each other, enabling them to learn the underlying distribution of surface roughness and aid in prediction tasks ^[62].

In a recent paper ^[63], a novel image processing method was proposed to enhance the quality of thermal images for feature extraction in the Directed Energy Deposition (DED)-based additive manufacturing process. The method utilized an Improved Enhanced Generative Adversarial Network (IEGAN) with a modified objective function. A penalty term was introduced to enhance the contrast ratio of the reconstructed thermal images. The effectiveness of the proposed IEGAN was demonstrated by comparing the contrast ratio with that of the original GAN. The IEGAN successfully extracted the shape of the melt pool, contributing to process monitoring in additive manufacturing.

3.2.4. Autoencoders

Autoencoders are a type of unsupervised learning model used for dimensionality reduction and feature extraction. They encode and decode input data to learn compact representations of surface roughness profiles. These representations can be used in traditional machine learning models or for visualization purposes ^[64].

In the context of additive manufacturing, post-processing methods are commonly employed to address surface imperfections and bulk defects. Previous studies analyzed the effects of different peening-based treatments on the fatigue performance of laser powder bed fusion AlSi1Mg samples. The fracture surfaces of failed samples were further analyzed, and machine learning-based approaches were utilized to identify correlations between the residual stress, hardness, surface roughness, depth of the crack initiation site, and fatigue life of the post-treated samples. A deep neural network (DNN) and stacked autoencoder (SAE) were used to develop a machine learning model, with the SAE providing accurate predicted results. Parametric analyses and sensitivity analyses were performed to assess the importance of each input factor. The results showed that enhancing surface hardening, inducing higher compressive residual stresses, and reducing surface roughness led to deeper crack initiation sites and improved fatigue life ^[65].

3.2.5. Deep Reinforcement Learning

Deep Reinforcement Learning (DRL) is a combination of deep learning and reinforcement learning techniques. In the context of surface roughness prediction in additive manufacturing, DRL has emerged as a promising approach for optimizing the manufacturing process to achieve desired surface roughness outcomes. By employing DRL agents, which are intelligent systems, optimal control policies can be learned through interaction with the manufacturing environment and feedback on surface roughness performance ^[66].

4. Conclusions

The findings demonstrate that AI-based methods have shown promising results in predicting surface roughness, offering benefits such as cost reduction and time savings. However, certain limitations and challenges, including the availability of quality data and model interpretability, need to be addressed.

References

Andreoni, A.; Gregory, M. Why and how does manufacturing still matter: Old rationales, new realities. Rev. d’Economie Ind. 2013, 144, 21–57.
Helper, S.; Krueger, T.; Wial, H. Why does manufacturing matter? Which manufacturing matters? A policy framework. SSRN 2021.
Pereira, T.; Kennedy, J.V.; Potgieter, J. A comparison of traditional manufacturing vs additive manufacturing, the best method for the job. Procedia Manuf. 2019, 30, 11–18.
Baumung, W. Design of an architecture of a production planning and control system (ppc) for additive manufacturing (am). Lect. Notes Bus. Inf. Process. 2020, 389, 391–402.
Framinan, J.M.; Perez-Gonzalez, P.; Fernandez-Viagas, V. An overview on the use of operations research in additive manufacturing. Ann. Oper. Res. 2023, 322, 5–40.
Gardan, J. Additive manufacturing technologies: State of the art and trends. Int. J. Prod. Res. 2016, 54, 3118–3132.
Gebhardt, A. Understanding Additive Manufacturing. Underst. Addit. Manuf. 2011, 2011, I–IX.
Rathee, S.; Srivastava, M.; Maheshwari, S.; Kundra, T.K.; Siddiquee, A.N. Friction Based Additive Manufacturing Technologies: Principles for Building in Solid State, Benefits, Limitations, and Applications; CRC Press: Boca Raton, FL, USA, 2018.
ASTM52900; Additive Manufacturing—General Principles Terminology. ISO: Geneva, Switzerland; ASTM: West Conshohocken, PA, USA, 2013.
Srivastava, M.; Rathee, S.; Patel, V.; Kumar, A.; Koppad, P.G. A review of various materials for additive manufacturing: Recent trends and processing issues. J. Mater. Res. Technol. 2022, 21, 2612–2641.
So, M.S.; Seo, G.J.; Kim, D.B.; Shin, J.H. Prediction of Metal Additively Manufactured Surface Roughness Using Deep Neural Network. Sensors 2022, 22, 7955.
Zhou, Y.; Chen, H.; Tang, Y.; Gopinath, S.; Xu, X.; Zhao, Y.F. Simulation and optimization framework for additive manufacturing processes. In Proceedings of the 2014 International Conference on Innovative Design and Manufacturing (ICIDM), Montréal, QC, Canada, 13–15 August 2014; Volume 17, pp. 34–40.
Dinovitzer, M.; Chen, X.; Laliberte, J.; Huang, X.; Frei, H. Effect of wire and arc additive manufacturing (WAAM) process parameters on bead geometry and microstructure. Addit. Manuf. 2019, 26, 138–146.
Le, V.T.; Mai, D.S.; Tran, V.C.; Doan, T.K. Additive Manufacturing of Thin-Wall Steel Parts by Gas Metal Arc Welding Robot: The Surface Roughness, Microstructures and Mechanical Properties. Intell. Syst. Ref. Libr. 2021, 193, 43–54.
Turner, B.N.; Gold, S.A. A review of melt extrusion additive manufacturing processes: II. Materials, dimensional accuracy, and surface roughness. Rapid Prototyp. J. 2015, 21, 250–261.
Strano, G.; Hao, L.; Everson, R.M.; Evans, K.E. Surface roughness analysis, modelling and prediction in selective laser melting. J. Mater. Process. Technol. 2013, 213, 589–597.
Li, Z.; Zhang, Z.; Shi, J.; Wu, D. Prediction of surface roughness in extrusion-based additive manufacturing with machine learning. Robot. Comput. Integr. Manuf. 2019, 57, 488–495.
Chan, K.S.; Koike, M.; Mason, R.L.; Okabe, T. Fatigue life of titanium alloys fabricated by additive layer manufacturing techniques for dental implants. Metall. Mater. Trans. A Phys. Metall. Mater. Sci. 2013, 44, 1010–1022.
Calignano, F.; Manfredi, D.; Ambrosio, E.P.; Iuliano, L.; Fino, P. Influence of process parameters on surface roughness of aluminum parts produced by DMLS. Int. J. Adv. Manuf. Technol. 2013, 67, 2743–2751.
Gülcan, O.; Günaydın, K.; Tamer, A. The state of the art of material jetting—A critical review. Polymers 2021, 13, 2829.
Wu, D.; Wei, Y.; Terpenny, J. Predictive modelling of surface roughness in fused deposition modelling using data fusion. Int. J. Prod. Res. 2019, 57, 3992–4006.
Du, Y.; Mukherjee, T.; Finch, N.; De, A.; DebRoy, T. High-throughput screening of surface roughness during additive manufacturing. J. Manuf. Process. 2022, 81, 65–77.
Lin, W.J.; Lo, S.H.; Young, H.T.; Hung, C.L. Evaluation of deep learning neural networks for surface roughness prediction using vibration signal analysis. Appl. Sci. 2019, 9, 1462.
Cao, L.; Li, J.; Hu, J.; Liu, H.; Wu, Y.; Zhou, Q. Optimization of surface roughness and dimensional accuracy in LPBF additive manufacturing. Opt. Laser Technol. 2021, 142, 107246.
Martín Fernández, F.; Martín Sánchez, M.J. Analysis of the Effect of the Surface Inclination Angle on the Roughness of Polymeric Parts Obtained with Fused Filament Fabrication Technology. Polymers 2023, 15, 585.
Mishra, P.; Sood, S.; Bharadwaj, V.; Aggarwal, A.; Khanna, P. Parametric Modeling and Optimization of Dimensional Error and Surface Roughness of Fused Deposition Modeling Printed Polyethylene Terephthalate Glycol Parts. Polymers 2023, 15, 546.
Kopp, M.; Uhlmann, E. Prediction of the Roughness Reduction in Centrifugal Disc Finishing of Additive Manufactured Parts Based on Discrete Element Method. Machines 2022, 10, 1151.
Saxena, A.; Murtaza, Q.; Kumar, P. Prediction of surface roughness in additively manufactured samples in PLA + polymer material through machine learning. Eur. Chem. Bull. 2023, 12, 2113–2124.
Mishra, A.; Jatti, V.S. Quantum Machine Learning Approach for the Prediction of Surface Roughness in Additive Manufactured Specimens. April 2023. Available online: https://arxiv.org/abs/2304.13142v1 (accessed on 10 August 2023).
Collins, C.; Dennehy, D.; Conboy, K.; Mikalef, P. Artificial intelligence in information systems research: A systematic literature review and research agenda. Int. J. Inf. Manag. 2021, 60, 102383.
Haroon, S.; Corien, P.; Erik, S. Mission AI; Springer International Publishing: Cham, Switzerland, 2023.
Murphy, R.R. Introduction to AI Robotics; MIT Press: Cambridge, MA, USA, 2014; pp. 1–487.
Liu, Y.; Zhao, T.; Ju, W.; Shi, S. Materials discovery and design using machine learning. J. Mater. 2017, 3, 159–177.
Du, S.; Liu, C.; Huang, D. A shearlet-based separation method of 3D engineering surface using high definition metrology. Precis. Eng. 2015, 40, 55–73.
Syamsuddin. Formation and improvement of surface waviness for additive manufacturing 5A06 aluminium alloy component with GTAW system. Rapid Prototyp. J. 2019, 32, 91–100.
Golhin, A.P.; Tonello, R.; Frisvad, J.R.; Grammatikos, S.; Strandlie, A. Surface Roughness of As-Printed Polymers: A Comprehensive Review. Int. J. Adv. Manuf. 2023, 127, 987–1043.
Wi, K.; Suresh, V.; Wang, K.; Li, B.; Qin, H. Quantifying quality of 3D printed clay objects using a 3D structured light scanning system. Addit. Manuf. 2020, 32, 100987.
Triantaphyllou, A.; Giusca, C.L.; Macaulay, G.D.; Roerig, F.; Hoebel, M.; Leach, R.K.; Tomita, B.; Milne, K.A. Surface texture measurement for additive manufacturing. Surf. Topogr. Metrol. Prop. 2015, 3, 024002.
Suh, C.H.; Jung, Y.C.; Kim, Y.S. Effects of thickness and surface roughness on mechanical properties of aluminum sheets. J. Mech. Sci. Technol. 2010, 24, 2091–2098.
Wu, D.; Wei, Y.; Terpenny, J. Surface roughness prediction in additive manufacturing using machine learning. In Proceedings of the ASME 2018 13th International Manufacturing Science and Engineering Conference, MSEC 2018, College Station, TX, USA, 18–22 June 2018; Volume 3, pp. 1–6.
Silbernagel, C.; Aremu, A.; Ashcroft, I. Using machine learning to aid in the parameter optimisation process for metal-based additive manufacturing. Rapid Prototyp. J. 2020, 26, 625–637.
Sendino, S.; Martinez, S.; Lartategui, F.; Gardon, M.; Lamikiz, A.; Gonzalez, J.J. Effect of powder particle size distribution on the surface finish of components manufactured by laser powder bed fusion. Int. J. Adv. Manuf. Technol. 2023, 124, 789–799.
Delfs, P.; Tows, M.; Schmid, H.J. Optimized build orientation of additive manufactured parts for improved surface quality and build time. Addit. Manuf. 2016, 12, 314–320.
Galati, M.; Rizza, G.; Defanti, S.; Denti, L. Surface roughness prediction model for Electron Beam Melting (EBM) processing Ti6Al4V. Precis. Eng. 2021, 69, 19–28.
Kumar, P.; Jain, N.K. Surface roughness prediction in micro-plasma transferred arc metal additive manufacturing process using K-nearest neighbors algorithm. Int. J. Adv. Manuf. Technol. 2022, 119, 2985–2997.
Barari, A.; Kishawy, H.A.; Kaji, F.; Elbestawi, M.A. On the surface quality of additive manufactured parts. Int. J. Adv. Manuf. Technol. 2017, 89, 1969–1974.
Campbell, C.; Ying, Y. Learning with Support Vector Machines; Springer International Publishing: Berlin/Heidelberg, Germany, 2011.
Singh, T.; Kumar, P.; Misra, J.P. Surface roughness prediction modelling for wedm of aa6063 using support vector machine technique. Mater. Sci. Forum 2019, 969, 607–612.
Jin, Z.; Shang, J.; Zhu, Q.; Ling, C.; Xie, W.; Qiang, B. RFRSF: Employee Turnover Prediction Based on Random Forests and Survival Analysis. In Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Springer International Publishing: Berlin/Heidelberg, Germany, 2020; Volume 12343, pp. 503–515.
Taunk, K.; De, S.; Verma, S.; Swetapadma, A. A brief review of nearest neighbor algorithm for learning and classification. In Proceedings of the 2019 International Conference on Intelligent Computing and Control Systems, ICCS 2019, Madurai, India, 15–17 May 2019; pp. 1255–1260.
Dastres, R.; Soori, M. Artificial Neural Network Systems. Int. J. Imaging Robot. 2021, 2021, 13–25. Available online: https://hal.science/hal-03349542 (accessed on 10 August 2023).
Soler, D.; Telleria, M.; García-Blanco, M.B.; Espinosa, E.; Cuesta, M.; Arrazola, P.J. Prediction of Surface Roughness of SLM Built Parts after Finishing Processes Using an Artificial Neural Network. J. Manuf. Mater. Process. 2022, 6, 82.
Boschetto, A.; Giordano, V.; Veniali, F. Surface roughness prediction in fused deposition modelling by neural networks. Int. J. Adv. Manuf. Technol. 2013, 67, 2727–2742.
Wafa, F.; Abdulshahed, A.M. Prediction of surface roughness in additive manufacturing using artificial neural networks. In Proceedings of the International Engineering Conference (IEC-MU22), Misurata, Libya, 6 October 2022.
Chen, H.; Yaseer, A.; Zhang, Y. Top Surface Roughness Modeling for Robotic Wire Arc Additive Manufacturing. J. Manuf. Mater. Process. 2022, 6, 39.
Roberts, D.A.; Yaida, S.; Hanin, B. The Principles of Deep Learning Theory; Cambridge University Press: Cambridge, UK, 2022.
Janiesch, C.; Zschech, P.; Heinrich, K. Machine learning and deep learning. Electron. Mark. 2021, 31, 685–695.
Saxena, A. An Introduction to Convolutional Neural Networks. Int. J. Res. Appl. Sci. Eng. Technol. 2022, 10, 943–947.
Abhilash, P.M.; Ahmed, A. Convolutional neural network–based classification for improving the surface quality of metal additive manufactured components. Int. J. Adv. Manuf. Technol. 2023, 126, 3873–3885.
Montavon, G.; Samek, W.; Müller, K.R. Methods for interpreting and understanding deep neural networks. Digit. Signal Process. A Rev. J. 2018, 73, 1–15.
Samek, W.; Binder, A.; Montavon, G.; Lapuschkin, S.; Müller, K.R. Evaluating the visualization of what a deep neural network has learned. IEEE Trans. Neural Netw. Learn. Syst. 2017, 28, 2660–2673.
Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative adversarial networks. Commun. ACM 2020, 63, 139–144.
Liu, W.; Wang, Z.; Tian, L.; Lauria, S.; Liu, X. Melt pool segmentation for additive manufacturing: A generative adversarial network approach. Comput. Electr. Eng. 2021, 92, 107183.
Michelucci, U. An Introduction to Autoencoders. arXiv 2022, arXiv:2201.03898.
Maleki, E.; Bagherifard, S.; Guagliano, M. Correlation of residual stress, hardness and surface roughness with crack initiation and fatigue strength of surface treated additive manufactured AlSi10Mg: Experimental and machine learning approaches. J. Mater. Res. Technol. 2023, 24, 3265–3283.
Plaat, A. Deep Reinforcement Learning; Springer Nature: Singapore, 2022.

© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.

Upload a video for this entry

Information

Subjects: Engineering, Aerospace; Engineering, Mechanical; Engineering, Manufacturing

Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to https://encyclopedia.pub/register :

Temesgen Batu

Hirpa G. Lemu

Hailu Shimels

View Times: 289

Update Date: 23 Oct 2023

Table of Contents

Video Upload Options

Confirm