Generalized Design for Additive Manufacturing (DfAM) Expert System Using Compliance and Design Rules

Abstract

Additive manufacturing (AM) has revolutionized the design and production of complex geometries by offering unprecedented creative freedom over traditional manufacturing. Despite its growing prominence, AM lacks automated and standardized design rules tailored to specific AM processes, resulting in time-consuming and expert-dependent manual verification. To address these limitations, this research introduces a novel design for additive manufacturing (DfAM) framework consisting of two complementary models designed to automate the design process. The first model, based on a decision tree algorithm, evaluates part compliance with established AM design rules. A modified J48 classifier was implemented to enhance data mining accuracy by achieving a 91.25% classification performance accuracy. This model systematically assesses whether input part characteristics meet AM processing standards, thereby providing a robust tool for verifying design rules. The second model features an AM design rule engine developed with a Python-based graphical user interface (GUI). This engine generates specific recommendations for design adjustments based on part characteristics and machine compatibility, offering a user-friendly approach for identifying potential design issues and ensuring DfAM compliance. By linking part specifications to various AM techniques, this model supports both researchers and engineers in anticipating and mitigating design flaws. Overall, this research establishes a foundation for a comprehensive DfAM expert system.

1. Introduction

Additive manufacturing (AM) involves the fabrication of intricate three-dimensional objects through a layer-by-layer technique with materials such as plastics, ceramics, or metals [1,2,3]. Traditional methods of manufacturing are limited in producing complex shapes and require extra steps, tools, and skilled labor, which raises the cost. In contrast, AM makes manufacturing complex geometries much easier with very few additional concerns [4,5]. Different types of AM processes include selective laser sintering (SLS) [6], fused deposition modelling (FDM) [7], stereolithography (SLA) [8], 3D printing (3DP) [9], and laminated object manufacturing (LOM) [10]. Expert systems, including artificial intelligence (AI)-based AM solutions, are employed in many domains, including process planning and work scheduling [11,12,13]. Three main factors impact an expert’s system job speed: (a) the geometrical complexity of parts, which in turn affects the yield of working space available around the part; (b) the total yield of the working space for this particular AM system; and (c) the technology used in the AM system [14,15].

AM processes have applications in various areas, including biomedical devices [16,17], semiconductor electronics [9], energy devices [18], and building and construction [19]. In each AM technology, there are distinct process capabilities, and the optimum use of material and machine resources requires expert assessment [20]. The large learning curve and specialized knowledge required for these technologies, however, do not allow for easy access to the AM technologies [21,22]. Biases of expert opinions and very limited off-the-shelf process guidelines add to the constraining factors against the spread of AM. Thereby, a generic system to assist novice designers and any AM enthusiast will serve as an important development toward its wider adoption in various applications [23,24].

Various research efforts have been made by both academia and industry to develop an expert AM-based system [25,26,27,28,29,30]. Kumke et al. illustrated a design concept framework to support design engineers during all phases of product development, enabling the full exploitation of potential from additive manufacturing and the development of AM-conformal designs [31]. Glen et al. investigated the additive manufacturing machine learning challenges by studying design repositories in a series of interconnected studies. A methodology demonstrated the study of the utility of large-scale engineering design repositories for additive manufacturing machine learning in both contexts: generalized and application-specific [32]. Stolt et al. showed how the integration of the AM design of components can be carried out with the design of the complete engine structure [33]. Wiberg et al. presented a DA (design automation) framework that smoothened the entire process of DfAM. It provides an overall view of the entire DfAM process by bringing on board all aspects concerning functional requirements to manufacturing evaluation and preparation within one design automation framework [34]. Pradel et al. proposed a DfAM framework that was based on a generic design process model made of five parts: conceptual design, embodiment design, detail design, process planning, and process selection [35]. Yao et al. introduced a hybrid machine learning algorithm for additive manufacturing design feature recommendation at a time during the phase of conceptual design [36]. Ghiasian et al. demonstrated an additive manufacturing recommender system for component inventories enabled by machine learning [37]. Ghazy et al. proposed an updatable decision support system, which supports a user’s choices of AM process chains, materials, finishing methods, and machines [38]. Yang et al. introduced an expert system for use in analyzing printability and integrability of assembly structure in design for additive manufacturing [39].

While previous efforts have been dedicated to automating the design and process planning stages of AM, there is still a need for a flexible framework that identifies AM process selections based on specific part requirements. In this research study, a data-driven system was developed for the optimum selection of AM processes based on CAD data and user-specified preferences. The AM techniques considered in this research include SLS, FDM, SLA, 3DP, and LOM. The specific objectives of the research included (a) generating a meta-model from an input part design to ensure conformity with the AM process rules and (b) developing a design for an AM rule engine, which gives recommendations on characteristics of the input part and compatibility with various machines.

2. Methodology

2.1. Design for Additive Manufacturing Framework (DFAM)

The design framework consists of four segments that include input (digital CAD design and extracting CAD features), database (AM process features thresholds and design rules guidelines, knowledge base, literature review, OEM database, and expert opinion), process analysis (decision tree algorithm and design rule engine), and output (DFAM rules compliance and rules recommendation). The overall framework of the research is illustrated in Figure 1.

2.2. Input CAD Feature Extraction

The digital computer-aided design (CAD) of the input part was exported as an STL file but can also be exported in STEP, IGES, OBJ, DWG, and DXF formats. The STL file was uploaded to the Magics software to extract crucial features such as minimum wall thickness, feature size, bosses/cylinders, hole size, minimum text size, tolerance, embosses, and engraved details, as shown in Figure 2.

Following the feature extraction, all gathered information was applied to a decision tree algorithm for evaluation. The decision tree requires a data sheet containing instances, where each instance is represented by features that describe the model to be 3D printed. The desired features—minimum wall thickness, hole size, embossed and engraved details, bosses/cylinders, minimum text size, feature size, and tolerances—were initially organized in a data file, with each row corresponding to a distinct part, as shown in

Table 1. Sample CAD features data sheet for 400 parts.

Part	Technique	Thickness (mm)	Hole Size (mm)	Embossed (mm)	Bosses (mm)	Text (mm)	Feature Size (mm)	Tolerance (mm)
1	SLS	1	1.5	1	2	0	0	1
2	SLS	0.5	2	0	1	0	0.8	0.6
3	FDM	0.5	0.5	0.6	0.5	1.5	0.4	0.1
4	FDM	0.8	1	0	2	4	1	0.2
5	SLA	0.5	2	0.2	0	0	0.4	0.5
6	SLA	3	0	0.5	0.5	1	0	0.5
7	3DP	5	5	0	5	0	0.2	0.2
8	3DP	15	0	0	0	0	0.2	0.1
9	3DP	2	2	0.1	0.5	0	0	0.2
..	SLS	20	10	0	5	1	0.4	1.5
..	FDM	7	0	1	0	4	0	0.6
..	SLA	0.3	0	1.5	0	0	0.8	1.2
400	SLS	0.4	0	0	0	2.2	0	1.9

. The AM techniques corresponding to respective 3D CAD parts are provided in Table 1 based on the DfAM framework established in this research study. Five AM techniques, which include SLS, FDM, SLA, 3DP, and LOM, were chosen for the evaluation. Each AM technique had 80 data instances derived from AM part sources [40]. Thus, our methodology can be applied to a variety of common input files used in the industry. Further, each type of input file can also be converted into STL format and vice versa, making it versatile in extracting the CAD features. In Figure 2, we demonstrate how an input part file can be used to extract dimensional information, which is then used to populate the datasheet shown in Table 1. The seven features, namely thickness, hole size, embossed details, bosses, text, feature size, and tolerance, are the most prominent features that are used in the literature to evaluate parts for different AM processes. Thus, our DfMA system provides a first-pass or first-stage methodology to seek if different parts can be manufactured with a variety of AM processes.

2.3. Database

The DfAM database developed in this research involved compilation of information from different sources including datasheets from OEM and expert opinions [41,42,43]. Crucial for harnessing the full potential of AM are process feature thresholds and guidelines on design rules that enable an effective DfAM framework [44]. The thresholds define the minimum wall thickness, hole size, feature size, and levels of tolerance among other factors, as shown in

Table 2. Threshold values for different features in AM processes.

No.	Feature Type	SLS	FDM	SLA	3DP	LOM
1	Min. Wall Thickness (mm)	0.7	1	0.8	1.2	1.8
2	Min. Hole Size (mm)	0.8	0.8	0.5	0.5	2
3	Embossed or Engraved (mm)	0.8	0.8	0.5	0.5	2
4	Bosses/Cylinder (mm)	0.8	0.8	0.4	0.5	2
5	Min. Text Size (mm)	2	2.5	1	1.5	3
6	Min Feature Size (mm)	0.5	0.5	0.5	0.5	1.2
7	Tolerance (mm)	0.3	0.5	0.2	0.1	0.8

[45,46]. Knowledge of these thresholds enables designers to make decisions regarding the feasibility of designs [47,48]. Violation of one or more of these process feature thresholds for an input part resulted in different compliance values, as explained in Section 2.5.

2.4. Process Analysis

Decision Tree Algorithm and Design Rule Engine: A model of the DfAM methodology consisted of four components, namely input part characteristics, database, process analysis, and output DfAM rules recommendations. This research work deals with efficient data mining procedures for predicting compliance with DFAM rules for an input part corresponding to each AM process [14]. The modified J-48 decision tree classifier algorithm was implemented [49,50,51] using the CAD features datasheet readable by the data mining tool WEKA [52,53]. By employing the meta model with data from both part characteristics and knowledge base, a program was developed using Python language to execute the design rule engine (Python) that develops DfAM rules recommendations [54]. The software requires only the part number as input; based on this, it searches for relevant constraints and provides appropriate DfMA (design for manufacturing and assembly) rule recommendations. After that, it performs the matching of input part number with relevant constraints available in the system and calculates the compliance percentage. In addition to compliance, the system finds both compliant and non-compliant design rules and generates recommendations. The number of recommendations is linked to accepted design rules, with an alert mechanism for instances falling below a minimum threshold.

2.5. Output

2.5.1. Compliance with DFAM Rules

Table 3. An expert judgment for compliance with DFAM rules.

No.	Compliance with DfMA Rules	% Ranges
1	High Compliance	76–100%
2	Moderate Compliance	51–75%
3	Low Compliance	26–50%
4	Poor Compliance	0–25%

shows the compliance with DfAM rules ranging from 0 to 100%. This table serves as the key concept in determining the compliance of the CAD design features dataset with respective AM processes.

2.5.2. Design Rules Recommendations

Table 4. Design Rules Recommendations for FDM.

Features	Threshold Value (mm)	Recommendations
Min. Thickness (mm)	≥1	Wall thickness should be large enough to structure each part layer with a boundary line and enclosed raster lines to minimize dimensional deviations and to avoid defects. If a wall is connected to another wall on two or more sides, it is advised to use a minimum wall thickness of 1 mm; if it is attached to another wall on just one side, it should be 2 mm.
Min. Hole Size (mm)	≥0.8	It is advised that the FDM’s holes be larger than 1 mm in diameter in order to maintain its circular shape. Holes must be oriented precisely, and printing parallel to the xy-axis usually yields the best resolution.
Embossed or Engraved (mm)	≥0.8	The minimum line thickness and depth for embossed and engraved elements should be 1 mm for the top and bottom of the design and 2 mm for vertical walls.
Bosses/Cylinder (mm) Like SLS	≥0.8	Cylinders outer radius should be large enough to structure each part layer with a boundary line and enclosed raster lines to minimize dimensional deviations and to avoid defects.
Min. Text Size (mm)	≥2.5	For protruding text to register correctly, it is advised that it maintain a minimum thickness of 0.04″ (1 mm). Text must also be at least 0.04″ (1 mm) tall, although for legibility and to prevent unforeseen errors, it is advised that it be between 0.047″ and 0.06″ (1.2 and 1.5 mm).
Min Feature Size	≥0.5	In order to print on general-purpose plastics, the minimum feature size is 2 mm. Choose a thickness of 1 mm if the feature is a thin wire that is joined on both sides; if not, we advise at least 2 mm.
Tolerance (mm)	≥0.5	Recommended tolerance of +/−1 mm.

shows design rules recommendations for the FDM technique. Further, Appendix A, Appendix B, Appendix C and Appendix D show the design rules for other AM processes that were used in this research. All information was gathered from a literature review, operation equipment manufacturing (OEM), and expert opinion.

3. Results and Discussion

The feasibility of the proposed DfAM framework was demonstrated in two stages. Stage 1 involved evaluating compliance with AM rules, and Stage 2 involved developing AM design rules recommendations.

3.1. Stage 1: Evaluating Compliance with AM Rules

In this stage, multiple case studies with different dataset sizes of 100, 200, and 400 were evaluated. The training to the test dataset was in a 80:20 ratio for all the above datasets. Thus, for the 400 dataset, there were 320 sample (parts) for training and 80 for the test dataset, respectively. The CAD model datasets were analyzed by the J-48 decision tree classification algorithm. The classification model was used to evaluate expert judgment classifying compliance with DfAM rules into four categories: highly, moderately, lowly, and poorly compliant [18]. In the table, design rules for compliance are highlighted with green and non-compliance with red. Compliance percentage was calculated based on Equation (1). Table 4 shows the minimum constraint values or threshold for each feature with respect to the FDM process. Thus, an actual CAD feature of the AM processes should be greater or equal to be acceptable or compliant for each feature [19].

Table 5. Sample datasets for AM process compliance of different CAD models.

No.	Process	Min. Wall Thickness (mm)	Min. Hole Size (mm)	Embossed or Engraved (mm)	Bosses /Cylinder (mm)	Min. Text Size (mm)	Min Feature Size (mm)	Tolerance (mm)	Expert Judgement
1	SLS	1	1.5	1	2	0	0	1	High Compliance
2	SLS	1.1	3	6	2.9	1	0.3	0.1	Moderate Compliance
3	SLS	0.5	2	0.2	0	0	0.4	0.5	Poor Compliance
4	FDM	0	0	0.9	1	4.1	2.3	1.7	High Compliance
5	FDM	0.6	1.4	2.8	2	1.7	0.4	0.3	Moderate Compliance
6	FDM	0.4	2.1	0.6	0.2	1.1	0.2	0.3	Poor Compliance
7	SLA	0	4.9	0	3	0	2.4	0.9	High Compliance
8	SLA	0.5	0.2	0	2	0	2	0	Moderate Compliance
9	SLA	5.3	0.2	0.3	0.3	0.6	0.3	0	Poor Compliance
10	3DP	1.9	0	0.7	0	0	0	0	High Compliance
11	3DP	0	1	0	0.3	0	0.3	0.5	Moderate Compliance
12	3DP	0.6	0.1	0	0.2	0.9	0.3	1	Poor Compliance
13	LOM	3.5	0	0	0	0	0	0	High Compliance
14	LOM	0.5	6	0.55	3	1.76	2.3	0.6	Moderate Compliance
15	LOM	0.4	0	0.7	0	6	0	0.15	Poor Compliance

Note: green for acceptable feature, red for unacceptable feature, and zero for non-applicable feature.

illustrates the sample datasets for AM process compliance of different CAD models.

$$Compliance percentage={\displaystyle \frac{(Acceptable Features\times 100)}{(Total Features-Not Applicapble Features)}}$$

(1)

Figure 3 illustrates the J-48 pruned tree classifier model as a full training test. Pruning reduces the complexity of the final classifier and hence improves predictive accuracy by the reduction of overfitting [55]. The main objective of this algorithm was to achieve perfect classification with a minimal number of decisions. In this research, the CAD dataset features of various AM processes such as SLS, FDM, 3DP, LOM, and SLA were investigated to determine compliance with AM design rules. In Figure 4, tree-based modeling is carried out with selective laser sintering (SLS) techniques in a classification framework. That framework depicts a multi-level decision tree with attributes and internal nodes, with the terminal nodes indicating predicted outcomes. This J-48 algorithm uses the information gain approach to obtain the best split. It returns information gained as the difference between the entropy before and after splitting. Entropy here is a measure for the impurity or uncertainty of information [56]. Equation (2) represents the formula of entropy.

Entropy = −pP × log2 (pP) − pN × log2 (pN)

(2)

wherein pP is the proportion of positive (training) examples, and pN is the proportion of negative (training) examples.

Case Study Results for 100, 200, and 400 Datasets

Initially, the training and testing datasets were imported into the WEKA software 4.2 to analyze the different types of datasets, such as 100, 200, and 400. The output accuracy for datasets 100 and 200 is presented in Figure 5 and Figure 6, respectively. The output accuracy for the 100 dataset was 51%, as shown in Figure 5. The output accuracy for the 200 dataset was around 66.5%, as shown in Figure 6, which is a significant improvement over the dataset with 100. The kappa statistic increased from 100 instances, and the error rates decreased. Each of the test instances was classified based on its respective eligible techniques, such as SLS, FDM, SLA, LOM, and 3DP. The lower right section of Figure 7 shows the count of expert judgment’s classification for each of the five AM processes.

Figure 7 illustrates the WEKA terminal with the 400 dataset for different AM techniques. SLS, the first bar, has an almost perfect distribution of the classification, with bluish green showing high compliance, sky blue showing moderate compliance, red showing poor compliance, and blue showing low compliance. FDM is falls majorly under moderate compliance, SLA shows fairly perfect distribution, 3DP is majorly classified as moderately compliant, and LOM similarly shows a majority of the instances being moderately compliant. With the 400 instance dataset, as shown in Figure 7, all the techniques had a fair spread of expert judgments. This was executed on WEKA, and the output accuracy was 91.25%, as shown in Figure 8.

From

Table 6. Details of accuracy by class for 400 datasets.

	TP Rate	FP Rate	Precision	Recall	F-Measure	ROC Area	Class
	0.956	0.055	0.837	0.956	0.892	0.972	High Compliance
	0.907	0.006	0.971	0.907	0.938	0.961	Moderate Compliance
	0.928	0.034	0.934	0.928	0.931	0.949	Low Compliance
	0.854	0.023	0.921	0.854	0.886	0.953	Poor Compliance
Weighted Avg.	0.913	0.031	0.916	0.913	0.913	0.957

, it can be illustrated that class-specific accuracy resulted in the highest ROC area of 0.972. The “Highly compliant” class had both the highest true-positive and false-positive rates, but the latter was contained within only 5%. The results indicate that further increasing the size of the train dataset and balancing the expert opinions about techniques make a real difference in additive manufacturing rules classification. In these methods, with statistical measures such as kappa statistics, error rates improved with the increase in the number of instances. The pruned tree created by the J-48 algorithm marked some important features decisive for any job compliance to actually be 3D-printed.

3.2. Stage 2: Developing AM Design Rules Recommendations

In this stage, the AM design rules were developed based on the range of compliance results. As mentioned earlier, a sample of 200 parts was considered, and the part quantity can be further extended based on the user requirements. To demonstrate the feasibility of the design engine rules, a case study on parts numbers 1, 50, and 80 was investigated.

3.2.1. Case Study for Part Number 80

Part number 80 was evaluated in the context of the DfAM rules recommendation framework. This part was related to the FDM AM technique. Based on the J48 classifier decision tree algorithm, the part compliance design rules were satisfied with a range of 75% only. Of the seven design features used, only four of them were considered for this part, and for one feature, the bosses did not meet the recommended minimum threshold constraint values for the features.

Table 7. Datasheet for part number 80.

No.	Tech	Thickness	Hole	Embossed	Bosses	Text	Feature Size	Tolerance	% Compliant	Expert Judgement
80	FDM	2	0	1.6	0.3	4	0	0	75	Moderate Compliance

Note: green for acceptable feature, red for unacceptable feature.

illustrates the datasheet for part number 80, the column: bosses were colored red if they missed the minimum threshold requirements. The design rule engine thus suggests recommendations for part number 80 on the bosses/cylinder feature. Figure 9a illustrates the CAD design of the part. Figure 9b shows the design rule engine window for part number 80. Finally, a new window containing recommended recommendations will open by clicking on the recommendations option in the design engine window, as shown in Figure 9c.

3.2.2. Case Study for Part Number 1

Part number 1 was evaluated by using the DfAM framework. Figure 10a represents the CAD design of the part and was related to the SLS technique. The J48 classifier decision tree algorithm suggested that all the compliance design rules were satisfied within the 100% range.

Table 8. Datasheet for Part number 1.

No.	Tech	Thickness	Hole	Embossed	Bosses	Text	Feature Size	Tolerance	% Compliant	Expert Judgement
1	SLS	2	0	1.6	0.3	0	0	0	100	High Compliance

Note: green for acceptable feature.

illustrates the datasheet for part number 1. Out of all the seven design features considered, only five of them applied to this part, and all those design rules met the recommended features’ minimum threshold constraint values. Thus, no recommendations were suggested by the design rule engine for part number 1. Figure 10b illustrates the design rule engine window for part number 1.

3.2.3. Case Study for Part Number 50

Part number 50, as shown in Figure 11a, was evaluated by using the DfAM framework as the SLS part. The J48 classifier decision tree algorithm analysis suggested that the design rules were poorly compliant, with less than a 25% range.

Table 9. Datasheet for part number 50.

No.	Tech	Thickness	Hole	Embossed	Bosses	Text	Feature Size	Tolerance	% Compliant	Expert Judgement
50	SLS	2	0	1.6	0.3	4	0	0	25	Poor Compliance

Note: red for unacceptable feature.

illustrates the datasheet for part number 50. Out of all the seven design features considered, four of them did not meet the recommended feature minimum threshold constraint values. Figure 11b illustrates the design rule engine window for part number 50. The findings represent that part number 50 results in multiple recommendations, as shown in Figure 11c, because the design rules did not meet the minimum threshold requirements of this object.

4. Discussion, Limitations, and Future Work

4.1. Discussion

The design for additive manufacturing (DfAM) framework engine developed in this research offers valuable utility to AM researchers and the industry, which relies on compliance with the rules of AM and a design rule engine. In the case of AM researchers, the DfAM framework engine acts as a meta-model in assessing compliance with AM processing rules. In this work, it is demonstrated that the framework diagnosed compliance with DfAM design rules for new datasets containing a total of 400 instances to give an appreciable classification accuracy of 91.25% using the tJ-48 decision tree of the WEKA data mining suite. This further enhances the engine capacity to scale up the training dataset to size, increases coverage of balanced expert judgments, and gives greatly improved classification results and hence useful insight and actionability for the researchers working on AM design [45]. The next best evidence of the effectiveness that the DfAM framework engine has to offer is in its ability to preempt or mitigate design problems in the paradigm [24,46]. The part numbers are mapped to the different additive manufacturing processes, and their associated constraints are retrieved while compliance is calculated, and the design rules are created. This feature allows spotting potential errors at the initial steps of the design process and thus helps to avoid costly iterations. The DfAM rule engine was developed in Python, which can be integrated easily into any existing workflow, and hence, its use by industry professionals will be easy and accessible [47]. The desired features considered in this research, which include minimum wall thickness, hole size, embossed and engraved details, bosses/cylinders, minimum text size, feature size, and tolerances, are the initial fundamental features for any AM process. These above-mentioned features of a 3D CAD design have a significant impact on the manufacturability, functionality, structural integrity, and postprocessing of the AM fabricated objects with respective AM techniques [23,24]. These features address the distinctive challenges of the respective AM processes, such as material behavior, geometric resolution limits, thermal effects, and postprocessing requirements. By precisely taking these features into consideration throughout the design phase, AM engineers can maximize the full potential of the producibility, cost efficiency, and performance of AM 3D components [25].

4.2. Limitations

Regardless of the developments in AM, there exist various shortcomings in design optimization. The current design guidelines, which include minimum wall thickness, hole size, embossed and engraved details, bosses/cylinders, minimum text size, feature size, and tolerances, are basic first-pass features. Despite the technological advancements in additive manufacturing, various limitations persist in terms of design optimization. Current design guidelines, including wall thickness, hole sizes, and tolerances, are often limited to specific materials or equipment corresponding to a respective AM process, which limits their general use [47,48]. To overcome such challenges, standardized, predictive, and sustainable design paradigms matched to the increasing capabilities of AM need to be established for broader adoption of the AM processes across different fields.

4.3. Future Work

Additive manufacturing design principles are evolving with new challenges and opportunities. Development of data-driven criteria for optimization of parameters such as build orientation, wall thickness and hole diameter, material, and the additive manufacturing process should be investigated to maximize the future prospects of AM processes’ manufacturability to full capability. Research on controlling material transition and optimization of tolerances in multi-material printing needs to be undertaken. Since the resulting parts can suffer from shrinkage and thermal stresses, enhanced dimensional accuracy is a requirement, which may improve using predictive models [24]. Design details such as text clarity need better geometrical designs or advanced layering techniques. AM tolerance practices need refinement for anisotropic properties and postprocessing. Standards should be established for new technologies, including hybrid AM, high-speed sintering, and minimum wall thickness. Integrated automatic tools and additive manufacturing constraints can aid in ensuring AM process manufacturability with maximized strength and functionality. This includes tools for analyzing lattice structures and providing guidance using topological optimization. Standardization of the additive manufacturing platforms will provide consistency in quality and ensure the full potential of the AM process for wide range of applications [46].

Future works can involve integrating machine learning algorithms within the design rule engine to enhance its capabilities and evolve the prediction-grade decision-support features of the DfAM framework. Future work may further populate the knowledge base through real-time data in the various AM processes and include feedback mechanisms to serve in further improving the adaptability and robustness of the design rule engine [57,58,59]. Through a more interactive user interface and making the engine compatible with different AM software, applicability and adoption could increase through the entire AM industry.

5. Conclusions

A design for additive manufacturing (DfAM) framework was developed in this research for various AM techniques by establishing compliance using decision tree algorithm and design rule engine recommendations. Crucial features such as minimum wall thickness, feature size, bosses/cylinders, hole size, minimum text size, tolerance, embosses, and engraved details were considered as baseline features to evaluate input parts for AM process compatibility. Threshold values for the above features were compiled using sources from the literature for five prominent AM processes commonly used in practice. These include SLS, FDM, SLA, 3D printing, and LOM. The compliance of the 3D parts was predicted by implementing the J-48 decision tree classification algorithm for multiple case studies with dataset sizes such as 100, 200, and 400 parts. The findings of the study demonstrated that the output accuracy of the 100 dataset and 200 dataset was 51% and 66.5%, respectively. This clearly indicates that data augmentation is vital for higher accuracies. Thus, an accuracy of 91.25% was achieved for AM compliance for a dataset of 400 parts. Moreover, the feasibility of the design rule engine developed by python graphical user interface was investigated by conducting case analysis on various 3D CAD parts designs. The Python-based engine enabled the early detection of potential errors that may be committed in the AM 3D design process, which is very promising for AM researchers and engineers to avert and minimize design errors. Three case studies were illustrated for distinct parts with 25%, 75%, and 100% compliance, along with their subsequent design modification recommendations. The results of the study illustrate the superiority of the DfAM framework not only in terms of providing accurate predictions of AM 3D CAD parts compliance but also demonstrating the recommended design rules for respective 3D CAD parts to maintain the AM design guidelines. The baseline feature considered in this research plays a significant role in evaluating the compatibility of a 3D CAD parts’ manufacturability with the five prominent AM processes, thus enabling the advancement of DfAM expert system development for the AM industry.