Cement-based additive manufacturing: experimental investigation of process quality

The construction sector accounts for almost 10% of the European gross domestic product (GDP) [1], and several European organisms and platforms aim to increase its innovation and efficiency [2]. The implementation of additive manufacturing (AM) is an important step toward achieving this purpose [3, 4]. AM is the process of using 3D model data to join materials, typically layer upon layer [5], for the production of objects. It allows the manufacturing of complex geometries with near-zero material waste; moreover, AM is applicable to a variety of materials, thus offering increased design freedom [6, 7]. The interest in AM has been steadily increasing in the last years [8]: it has been estimated that globally, the use of AM exclusively accounts for an added value of $667 million [9], a fact that is very encouraging concerning new ideas and adaptations of AM in other sectors. Liang et al. [10] highlighted the advantages of cement-based AM applications by using examples from 3D-printed buildings in China.

A large number of different AM processes are currently available [11]; these processes vary in terms of their process mechanism, the manner that the layers are deposited, and the materials that can be used [5, 12]. The vast majority of the existing studies and applications referring to cement-based AM [13] utilize a modified form of material extrusion; the use of AM processes other than cement-based ones is scarce [14]. Cement-based AM differs in three major aspects from AM applications in which other materials are used: the process mechanism, the materials, and the scale. The process mechanism is not thermal-based, as in the majority of AM applications [7]. The material properties and the difference of two orders of magnitude in scale (nozzle diameter, layer height) significantly affect the phenomena that take place. The application of cement-based AM requires studies that are focused on more efficient process parameter calibration and machine control. However, owing to the fact that AM for concrete-based materials for the construction sector is a rather new application, relevant studies are quite limited [13].

Most studies on cement-based AM and other AM applications for the construction sector have been focused on material-related issues. Such a representative study is that of Le et al. [15]; they studied the effects of different dosages of material components on the resulting shear strength. Moreover, they correlated the workability and open time of the final material to the aforementioned material dosages. Gosselin et al. [16] presented a procedure that enabled the modification of the material properties of the mix as it moved toward the nozzle. The same technique was investigated by Esnault et al. [17]; in their study, a special viscosity-increasing additive was injected to the mix before it would reach the nozzle. This technique improved the printability of the concrete materials, while simultaneously allowing them to retain their cohesion, which is an important challenge in cement-based AM [10].

Another group of studies has been focused on the development of AM machines. Hinczewski et al. [18] developed a stereolithography-based AM application for ceramic materials, which was used for the study of the effects of each material component. The AM machine developed by Khoshnevis et al. [19] was based on the contour crafting process; also, the influence of the design parameters on the product quality was investigated. Lim et al. [20] developed an automated concrete AM machine that was used for the investigation of material parameters and design aspects.

Studies focusing on other issues also exist, such as that of Salte et al. [21], in which the effects of the design, printing, and assembly on the structure of the final product were studied. In [22], how the tensile bond strength of printed mortar is affected by various process parameters was investigated. However, in cement-based AM, one of the crucial issues that needs to be addressed through research studies is that of higher precision and productivity [23]. Nevertheless, there is a shortage of studies focusing on such process-related issues.

In this work, the effects on quality of controlling the path width were investigated, using an alternative control strategy. The path width was controlled by changing the scanner head speed, while maintaining the extrusion speed (material flow) constant. This led to a change in the ratio of the aforementioned speeds, thus affecting the width of the deposited path. The advantage of this approach is its higher precision and productivity because the response time of the scanner head is faster than that of the extrusion head-system for large-scale cement-based AM applications [24] (Fig. 1). Furthermore, the effect of important process parameters on the part quality were studied; these process parameters were the extrusion speed, the ratio of the extrusion/scanner head speed, and the radius under which the extrusion takes place. Moreover, linear- and curved-extrusion experiments were conducted utilizing a small-scale experimental setup that had been designed and constructed in-house.

Two types of experiments were conducted: linear and curved extrusion. The selection of the process parameters, the key performance indicator (KPI), their levels, and the design of the experiments will be described in this section.

The process parameters that were selected for the linear-extrusion experiments were the extrusion speed and the ratio of the extrusion/scanner head speeds. The first process parameter was selected because it is currently used for the control of the material flow in most AM applications. The second process parameter was chosen in order to define the range of values that maintain the quality of the resulting parts. The process parameters used in the curved-extrusion experiments were the ratio of the extrusion/scanner head speeds and the radius under which the extrusion would take place. The extrusion/scanner head speeds, which will be hereafter referred to as ratio, are calculated via the following equation:where v_ex is the speed of the material when exiting the nozzle of the extruder, which will be referred to as extrusion speed, and v is the speed of the scanner head of the machine.

The levels of the process parameters of the linear experiment were selected in order to identify the ratio range that would lead to the production of successful parts. Smaller intervals were employed between the levels of the process parameter for the ratio in the range that was within the expected critical ratio threshold (i.e., levels 1, 2, and 3). Larger intervals were employed between levels 4 and 5. The rotational experiments were conducted using the ratio range that yielded successful linear parts in order to determine the significance of the effect of curvature. The behavior of the fresh concrete paste was viscous [25]; consequently, it had certain relaxation times [26]. In order to accurately capture a linear, a Kohlrausch–Williams–Watts (KWW), or even non-linear behavior [26], five different levels were used for the material extrusion variable [27‐30].

The KPI that was used as the response factor was the part consistency–quality metric (PCQM). The PCQM is a qualitative empirical metric, reflecting the consistency and quality of the extruded part. The higher the value of this metric is, the better the quality and the consistency of the extruded test part is. Five levels were used for the PCQM: levels 4 and 5 corresponded to high-quality parts, level 3 to acceptable ones, whereas levels 1 and 2 could be considered as non-acceptable. More specifically, level 1 has been used for parts that showed lack of continuity, such as holes and cavities, or a width variation of more than 40% along their length. Level 2 has been used for parts that might be continuous, but that had an uneven surface and a width variation along their length was within the range of 25–40%. Level 3 has been used for parts that had a width variation of 15–20%, featuring some mild inconsistencies and uneven surfaces. Finally, levels 4 and 5 for parts that were even in their width, with smooth and consistent surfaces. More details can be found in Table 1.

Test parts that presented lack of cohesion and continuity, as well as width inconsistencies and very uneven surfaces, have been characterized as non-acceptable. The reason for this was that a multilayer part consisting of layers with the aforementioned characteristics would be of low quality, or would even fail, owing to its poor dimensional accuracy, very uneven surface, and low mechanical properties, especially in the case of transverse loads. Images of the test parts (ribbons) of different PCQM levels can be seen in Fig. 2.

Tables 2 and 3 list the process parameters, which are the factors of the linear and the rotational experiments, as well as their corresponding levels and their values.

Owing to the fact that two factors were used in each experiment, an exhaustive combination is required according to the Taguchi factor analysis [31]. Consequently, 25 linear experiments and 9 rotational experiments were conducted. The combinations of the levels of factors employed in each experiment are summarized in Tables 4 and 5.

To analyze the results, an analysis of variance (ANOVA) was performed. The selected objective function (signal-to-noise (SN) ratio) was the-larger-the better because the larger the value of the PCQM would be, the better the quality and the consistency of the test part was [31]. The optimum level for a factor was the one that would yield the highest SN value within the experimental region. Figure 7 shows the plot for the SN ratios of the linear experiment. The optimum level values of the two factors are (i) extrusion/conveyor ratio = 1.2 (level 4) and (ii) extrusion speed = 12 mm/s and 15 mm/s (both levels 1 and 2). However, it may be observed that the effect of the different levels of factor B is not significant because the SN value is maintained fairly constant for the different values of extrusion speed.

A cross-section view and a top view of five experimental samples are illustrated in Figs. 8 and 9, respectively. The quality is the highest in test part (b) and is closely followed by that of (a); part (c) is still of acceptable quality. Parts (d) and (e) can be considered to be of unacceptable quality. More details about the process parameters and the PCQM of the test parts are listed in Table 9. By comparing the cross-sections from part (a) to part (e), it may be observed that the cross-section area decreases; this is attributed to the fact that the ratio from test part (a) to test part (e) decreases as well.

Considering the experimental results, the speed ratios of the extrusion speed and scanner head that lead to the production of test parts of acceptable quality are:

These results have also been validated through ANOVA (Table 10), in which the P value of factor A (ratio) was found to be significantly lower than the value used for α; this indicates a significant difference among the levels of this factor. On the contrary, the P value of factor B (extrusion speed) is significantly higher than the value used for α, indicating that there is no significant difference between the levels of this factor. This is reflected in the very unbalanced 95.67% and 0.91% contributions of factors A and B, respectively, which was calculated using the total sum of squares and the sum of squares due to the two factors (A, B). The sum of squares is equal to 4.19% owing to the presence of error. The results of the F test are equivalent as well.

The same procedure has been followed for the rotational tests as well. Figure 10 illustrates the plot for the S/N ratios. According to the plot of Fig. 10 and by selecting the level that corresponds to the maximum value of each factor, the optimum level values are (i) extrusion/rotary table ratio = 1.27 (level 3) and (ii) radius = 200 mm (level 3).

A cross-section view and a top view of four experimental samples are shown in Figs.11 and 12, respectively. The ratio used for the test parts decreases from test part (a) to test part (d); this is reflected in the area of their cross-section, which also decreases. The quality is the highest in test parts (a) and (b), followed by (c). Part (d) can be considered to be of unacceptable quality. More details about the process parameters and PCQM of the test parts are listed in Table 11.

These results have also been validated through ANOVA (Table 12), in which the P value of factor A was found to be significantly lower than the α value, which indicates a significant difference among the levels of this factor. The P value of factor B is significantly higher than that of factor A and is close to the value used for α; however, it is lower than the value used for α, indicating that there is a somewhat significant difference between the levels of this factor. The same is reflected in the F test.

Using the total sum of squares and the sum of squares due to the two factors (A, B), the responsibility (%) of each variation factor can be estimated. The extrusion/rotary table speed (factor A) is an important factor that significantly affects the quality and consistency of the test part (73.58% responsibility). The radius is also important (factor B), but not equally so (26.03% responsibility). The sum of squares due to error was 5.49%. The high significance of the speed ratio was expected, whereas the significance of the radius was approximately one-third of that of the speed ratio. Therefore, the ratio is the most important factor; however, the curvature also affects the part quality and cannot be neglected.

An important observation based on the curved-extrusion experiments is that the quality and consistency of the outer side of the part tended to be lower than those of the inner side of the part. This was attributed to the fact that the phenomenon of under-extrusion is more intense on the outer side of the part than on the centerline, and even less so on the inner side. Based on this observation, the following equation was developed:where ratio_loc is the local ratio, R_loc is the local radius, and ω is the rotational speed of the rotary table. Therefore, to calculate the ratio on the outer side of the part, ratio_outer, the R_loc value that corresponds to the outer side of the test part should be used.

To demonstrate the above, two cases are presented; more specifically, Fig. 13 illustrates the difference in quality and consistency between the inner and the outer side of different test parts. The ratio calculated on the centerline of part no. 2 is 0.8. However, the outer ratio of the part (which was calculated using Eq. (3)) is 0.75. This value is below the threshold of 0.8, which is the value that had been defined as the limit over which successful parts are created. More details on ratio calculations on the inner side, the outer side, and the centerline of the parts of Fig. 13 are summarized in Table 13.

This equation can provide a good indication on whether a part will be successful: by ensuring that the local ratio of the outer side of a part is higher than the critical ratio, successful curved parts of any radius can be manufactured.

In this experimental study, an experimental apparatus was designed and constructed to investigate the significance of important process parameters on the quality of parts manufactured via cement-based AM. Moreover, the Taguchi and ANOVA methods were applied. It was found that in linear extrusion, the part quality depended solely on the ratio of the extrusion/conveyor belt speeds. This was also indicated by the ANOVA results: 95.67% contribution for the extrusion/conveyor ratio and 0.91% for the extrusion speed. The value range of the ratios that yielded an acceptable part quality was 2 > ratio_acceptable > 0.8. Using ratios below 0.8 resulted in uneven surfaces because the material was stretched, thus being irregularly deformed. In curved extrusion, the effect of the ratio was the most important parameter (73.58%). However, the radius under which the extrusion took place also played an important role (26.03%). Consequently, in real applications, not only the ratio but also the curvature should be considered in the process of achieving acceptable part quality. However, it was observed that during curved extrusion, it was the outer side of the part that demonstrated issues in part quality and cohesion. Therefore, by calculating the ratio on the outer side of the part and by maintaining it within acceptable values (ratio_acceptable > 0.8), parts of acceptable quality and consistency can be created for any curvature. Thus, only one parameter has to be considered, namely the outer ratio, which simplifies the control procedure.

By following the proposed path-width control strategy, increased part quality can be achieved. The corresponding optimal parameters might differ for a different material mix; however, the same approach can be implemented to define them. Therefore, by identifying the value range of acceptable ratios for linear extrusion and by utilizing Eq. (3), the manufacturing of high-quality parts using different cement-based materials for both linear and curved paths can be achieved.