**1. Introduction**

Customer demands on products and services are constantly increasing in the global and local market and the competitive conditions of companies are in constant flux. As a result, companies are faced with changing market conditions in which they have to assert themselves against increased competitive pressure, greater product complexity, and an ever-increasing variety of products [1].

Industrial 3D printing or additive manufacturing (AM) is considered the key technology for mass customization. This allows the individual production of complex components and offers various possible solutions for increasingly complex requirements [2]. Additive manufacturing is an umbrella term for manufacturing processes in which components are built up element by element or layer by layer directly from computer-aided design (CAD) data without component-specific tools [3].

Fused deposition modeling (FDM) belongs to the AM technologies, which enable incorporation of cavities in a part's design and have little changeover cost compared to conventional manufacturing processes, potentially enabling individualized production and new possibilities for light-weight products [4].

The first step of state-of-the-art FDM processes involves a so-called slicing software that is used to generate machine-executable instructions (G-code), approximating a virtual product in the form of a CAD model. The desired product geometries are decomposed (sliced) into stacked layers of equal height along a specified axis, called the build orientation. Furthermore, for each of the layers, a closed two-dimensional path is planned, incorporating print head velocities set by the user. In the production step, the FDM machine follows the defined path while extruding heat-liquified raw material threads. Starting on the build plate, for each layer, the respective trace is followed by the machine, and so the whole part is fabricated. During the process, the machine must highly accurately control the correct material flow, the build plates state, and the correct positioning and velocity of the print head [5, 6].

While the concept seems straightforward to realize, practitioners long since report that reproducibility and reliability issues persist, demanding effective quality control measures [7–9]. Achievement of the aforementioned demand requires controlling for influences from the 5M-domains during pre-process and in-process stages: man, machine, milieu, material, and method affect the success of a print [10].

To achieve a successful production result, a set of numerous interrelated process parameters must be determined, some of which have been mentioned above. Finding appropriate parameters can pose a challenge to beginners, leading to failure in almost every second print [11]. Even expert knowledge does not necessarily lead to a good print result, but their experience helps them to avoid easily preventable mistakes. Many manhours and unnecessarily wasted material could be saved by the prevention of simple mistakes. The reason lies in the fact that there is almost no recognized or approved reference process in which defined requirements serve as a quality control measure.

As of now, there are only few works that serve as a reference process for additive manufacturing. There is a lack of standards and norms that ensure high process and product quality. A lot of previous academic literature focuses on particular printing defects such as warping or oozing. Performed research indicates that comprehensive guidelines regarding failure prevention in the overall printing process ought to be developed. Additionally, there is a lack of in-depth documented requirements to achieve high quality in process and printed products.

This paper proposes a reference process model including 10 quality gates that serve as documented requirements to prevent defects and failure prints beforehand instead of costly troubleshooting. Section 2 describes related work and shows up the gaps upon which this present scientific work further elaborates. In Section 3, various failure types are introduced and considerations to prevent them. Section 4 contains the proposed reference process model including quality gates. Section 5 summarizes the results and discusses the advantages of the proposed approach and concludes with future research potentials.

#### **2. Related work**

An extensive list of FDM print issues and their causes have been published by Loh et al. [12]. Qualitative expert knowledge has been formulated in natural language and

#### *Development of a Quality Gate Reference Model for FDM Processes DOI: http://dx.doi.org/10.5772/intechopen.104176*

lacks precise arguments. Each failure is assigned to a category which is either "printerassociated," "deposition-associated," or "print quality," but no reason nor meaning for an assignment is given. Livesu et al. provide a detailed description of the main process starting with the CAD model and ending in the G-code, but they focus mainly on software issues [13]. Baş et al. describe print conditions that sufficiently lead to faults by the application of a fault tree analysis (FTA) [14]. Many faults are described and their dependencies are formally expressed. The German Norm DIN SPEC 17071 proposes a print flowchart, leaving open the actual events of quality checks [15]. Oropallo et al. name error control in a list of 10 challenges in 3D printing. A distinction is made between errors during printing and errors before printing, which is partially avoidable [16]. Bähr and Westkämper divide a print into three stages: pre-process, in-process, and post-process. The importance of cooling is emphasized and divided up into a sinter phase, crystallization phase, glass transition phase, and a shrinkage phase, which are bounded by corresponding temperature values. Additionally, a table is provided that relates process parameters to their manifestation in component properties. Martinez-Marquez et al. developed a detailed quality control procedure including 18 quality gates but tailored to the production of patient-specific medical implants [17]. The process assumes the use of a laser-based AM system and error control is only briefly described. Fu et al. provide literature research and an overview of sensor technologies for in-situ monitoring of FDM processes [18]. Their flowchart is limited to in-situ printer health and product quality monitoring. Oleff et al. do systematic literature research in order to find quality-related research gaps, giving examples for a few FDM-process errors [19]. Song and Telenko examine FDM-print failures in a university makerspace [20]. They categorize these into user errors, machine errors, and designer errors. Also, a poll has been carried out to determine failure rates dependent on the user's experience level. The results show wastage levels of about 34% of the total material and a print failure rate of 41.1%. Gibson et al. provide a rough overview starting with the CAD model, ending in the application. An in-process view, as well as defects, is not considered.

To the best of the authors' knowledge, no publication at present exists wherein a generic reference process is determined in which quality gates serve as requirements for quality control to prevent printing defects.

### **3. Defects in additive manufacturing**

Defects that occur later in the process chain are harder to assess, as this presumes that no defect has occurred in a preceding process step.

In the following, examples of failures are explained which can either be pre-process, in-process, or post-process. Also, dependencies among failures are illustrated and research hypotheses for their assessment are formulated. Most of the enumerated defects have already been explained by Loh et al., whose work is extended in the following.

#### **3.1 Pre-process defects**

The following shows defects that may occur in the pre-process steps and are possibly preventable through quality control measures.

#### *3.1.1 Tangled filament*

If the end of the filament thread on a roll has been guided through under itself, a knot will eventually form on this roll, making proper unwinding impossible. This can happen after a user has unloaded filament from a printer. A proper loading process of filament should therefore be examined and verified.

#### *3.1.2 Gaps*

In all instances of this kind of defect, print segments are not properly connected and small gaps are recognizable by the naked eye. Loh et al. distinguish between three kinds "walls not touching," "gaps between Infill and outline" or "gaps between thin walls." Such gap appearances are introduced by the slicing software, affected by the extrusion line width. Thus, gap defects are avoidable if slicing errors are being determined.

#### *3.1.3 Small features not printed*

This defect highlights noticeable differences between the provided CAD model and the production instructions executed by the printer. Two distinctions between affected features can be made: A vertically standing wall whose width is smaller than the extrusion line width and a feature parallel to the build plate, whose height is smaller than the layer height. Material waste can be prevented if the slicing software informs the user about deviations between the CAD model and the generated G-code.

An example part that is susceptible to these kinds of defects is given in **Figure 1**, along with G-code paths generated by a slicing software using different parameter settings. The part consists of a block, a thin wall whose depth is 0.35 mm on its top, and a thin feature whose height is 0.1 mm in parallel to the build plate and is shown in (a). If a layer height smaller than the height of the thin feature of the CAD model and a line width smaller than the wall depth is chosen, then both the wall and the thin feature are not included in the generated G-code path, as demonstrated in (b). Conversely, if the layer height is smaller than the thinnest feature and the line width is smaller than the wall depth, then the sliced result matches the expectation of the

**Figure 1.** *Variations of parameter settings in slicing process of a complex cuboid object.*

user, illustrated in (e). If either layer height or line width is not chosen appropriately, corresponding results will be sliced, as can be seen in (c) and (d), respectively.
