2.2 Economic sustainability of additive manufacturing
2.2.1 Build time estimation models of additive manufacturing
Build time is defined as the time required to produce the part in the AM based machine.
Sometimes, time consumed in pre-processing operation and post-processing operation is also considered as a part of build time. Several attempts have been made to understand the effect of machine and geometry related parameters on build time of an AM process.
Estimation of an AM based technology is process specific and varies according to the working principle. Numerous models, viz., analytical, parametric and soft computing models have been proposed for estimating the build time in AM process. The following subsections discuss different build time estimation models developed in the past.
2.2.1.1 Analytical models
In one of the earliest investigations, Kechagias et al. (1997) attempted to estimate the build time for two different laser-based AM processes, i.e., stereolithography (SLA) and laminated object manufacturing (LOM). At that time, these technologies were at infant stage and the term ‘additive manufacturing’ (AM) was not coined. These processes were referred to as rapid prototyping (RP). For both the processes, time was estimated as algebraic summation of time taken in layer preparation and layer processing. Layer preparation comprises the time to form a new layer, time required by the platform to move downwards and delay time. Layer processing time comprises the time taken by the laser to scan a layer according to the geometry of the part. Different parts were fabricated to validate the accuracy of the proposed methodology. The deviation between the experimental value and estimated value was estimated. For SLA, the lowest and highest deviation of build time was found to be 0.3% and 14.5%, respectively; whereas for LOM, the lowest and highest deviation was found to be 3.45% and 54.83%, respectively.
In another early attempt, Alexander et al. (1998) proposed a methodology to estimate the build time of two popular AM processes, i.e., fused deposition modelling (FDM) and SLA. These processes were referred to as layered manufacturing. The build time (tbuild) was given by
b u ild m a n idle
,
t t t
(2.1) where tman is the manufacturing time where the part and support are produced in a layer by layer fashion. On the other hand, tidle is the non-production time. This time period involves the movement of the nozzle without depositing material, cleaning of the nozzle, command execution time and movement of the z axis. The time estimated by this method underestimated the true time, as the effect of acceleration, deceleration and the time taken for changing the direction of the nozzle was not considered.A popular laser-based process that uses polymeric material is selective laser sintering (SLS). It is one of the oldest and popular powder bed fusion AM processes.
Specific to SLS, Pham and Wang (2000) investigated the effect of machine parameters such as roller speed, roller travel distance, delay time, laser scan speed and laser scan spacing for estimating the build time. The study considered two SLS machines and four different raw materials for case studies. The maximum error, i.e., the difference between actual and estimated build time was 8%. The authors highlighted that inclusion of the effect of acceleration and deceleration of the roller can make the model more accurate. In another
study confined to SLS, Zhang and Bernard (2013) proposed a method to estimate the build time for a single part, similar multiple parts and mixed parts. The model comprised laser- based parameters such as laser velocity, laser diameter and hatching distance. However, some important parameters specific to SLS process such as roller velocity, roller travel distance, delay time was not considered. Instead, the recoating time, a part of layer preparation time was taken as six seconds. The time required for machine preparation and ending operation was considered as one hour. Two products having different geometrical features and sizes were considered for the case study; these were dissimilar products and the ratio of their volumes was 19:1.
In extrusion-based AM process, i.e, FDM, Thrimurthulu et al. (2004) proposed a methodology to estimate the build time of a part based on fundamental machine and geometrical parameters. They assumed that the deposition of the thermoplastic material is continuous and there is no interruption. Also, the study ignored the non-productive time and the effect of orientation of the part in the printing process. Assuming continuous deposition of the molten material, tbuild was obtained as
p
,
build
n e
t V
a v
(2.2) where Vp is the volume of the part to be fabricated, an is the cross-sectional area of the nozzle and ve is the extrusion velocity of the filament. The time estimated by this method will be different from the actual build time, as non-productive time (lowering down of the platform, time between ending and starting between two different paths) was not considered. This gap was addressed in the work of Han et al. (2003). They carried out a build time analysis for FDM process. The algorithm to estimate the build time comprised the procedure to determine the repositioning and cleaning time of the nozzle. The authors proposed adaptive slicing (variable layer thickness) method by dividing the part into different zones; each zone was assigned a different layer thickness. Parts having simple geometry, e.g., a cylinder, comprised only one zone. However, the parts having complex geometrical features like sharp curvatures had multiple zones.In another study specific to FDM process, Komines et al. (2018) presented analytical as well as empirical build time estimation models, which were validated by experiments. As per the authors, the build time is highly influenced by the acceleration of the nozzle especially for fabricating complex geometrical parts. In the analytical approach, the deposition time was estimated as the time required by the nozzle head to move with
constant velocity and the acceleration time of the nozzle. On the other hand, in the parametric approach, an acceleration coefficient was considered. However, the effect of different geometries on acceleration was not mentioned explicitly. The analytical model produced relatively more accurate results than the parametric model.
2.2.1.2 Parametric models
Specific to SLS, Ruffo et al. (2006b) proposed a parametric model for estimating the build time. They explored the effect of geometrical parameters on the build time of an object.
The overall build time was given by algebraic summation of recoating time, scanning time, and heating and cooling time (thc); thc was taken as one hour. Case studies were conducted and it was reported that two parts having same height and volume may take different time to build.
The effect of different orientations (rotation about different axes) of a part on the build time was explored by Rathee et al. (2017). They built a cylindrical shaped object by FDM process. They conducted several experimental runs and used response surface methodology, a statistical based technique, to establish parametric models. The layer thickness had the maximum effect on the build time irrespective of any orientation. Build time was the least for orientation having the minimum z height; although it may increase the cross-sectional area of a layer, it also reduces the number of layers in the z direction.
However, it was observed by Chacon et al. (2017) that specific to some geometries and process parameters, the orientation having lesser number of layers took relatively more time. High printing speed and feed rate may increase the deposition rate of the material, but repositioning of the nozzle takes additional time due to the effect of acceleration and deceleration.
2.2.1.3 Soft computing models
Some researchers used artificial neural network (ANN), a popular soft computing technique in build time estimation of AM processes. For example, Munguia et al. (2009) proposed an ANN model for the time-estimation in SLS process based on geometrical features of the product. They considered height of the part, volume of the part and bounding box as input parameters (referred to as input neurons). A bounding box is an imaginary cuboidal shaped box that contains every edge of the part. Munguia et al. (2009) took same examples as taken by Ruffo et al. (2006b) and compared the results. The average error in time estimation obtained by ANN model was reduced to 2.08% from 14.98% as obtained by empirical relation of Ruffo et al. (2006b). In another ANN based study, Di Angelo and Di Stefano
(2011) estimated the build time of several AM processes including SLA, SLS, FDM, LOM and three-dimensional printing (3DP). For this, they considered the part’s height and volume, layer thickness and repositioning movements. Experiments were conducted only for FDM and 3DP considering six test cases. The average percentage errors for FDM and 3DP were 11.5% and 12%, respectively.