Page 16 - ESAM-1-4

Engineering Science in
Additive Manufacturing Experimental statistics in AM

device-specific protocols (e.g., ASTM F382 for bone plates, ones (Figure 8A), compared to the more diverse options used
ASTM F2077 for intervertebral fusion devices, and ISO in PBF. While they do not use any problematic designs, such
7206 series for hip joint prostheses) define test methods as one factor at a time or single-value experiments, they also
for static and dynamic loading, fatigue life, and wear do not utilize any RSM designs, such as central composite
performance. 35,36 These standardized methods provide designs (CCD). A similar pattern holds for statistical analyses
clear procedures for specimen preparation, loading (Figure 8B). Most orthopedic studies use either an ANOVA
regimes, and acceptance criteria, ensuring reproducibility or t-test and are much less likely to just report descriptive
and comparability across laboratories. statistics; they also do not use more advanced models, such as
a regression. Finally, the proportion of papers that utilized at
Biological safety is addressed through the ISO 10993
series, which provides detailed protocols for evaluating least one useful DOE feature is higher in orthopedic studies
and PBF. This is primarily the case because most studies
cytotoxicity, sensitization, irritation, systemic toxicity, and mentioned that experimenters randomly assigned subjects
37
long-term effects of implantable materials. These tests to different treatment groups. Generally speaking, these
are essential for verifying that residual powders, post- patterns have not changed between 2016 and 2024, although
processing treatments, or unique surface morphologies the proportion of PBF papers using good experimental
generated during AM do not introduce biocompatibility strategies does increase (Figure 8C).
risks. Alongside mechanical and biological testing,
robust statistical analysis is required to substantiate 3.5. Best of the sampled papers
compliance claims. Standards such as ISO 16269 The use of statistics and DOE is not new to the field of
(statistical interpretation of data) and ASTM E2709/E2810 PBF-LB/M. Several manuscripts have successfully used
(statistical methods for demonstrating conformance with many of the techniques described in this review to address
specification limits) offer frameworks for determining empirical questions. Here, we highlight the best papers
sample sizes, confidence intervals, and equivalence we sampled that are focused on PBF-LB/M. These papers
margins. 38,39 Regulatory guidance, including the FDA’s were selected based on the number of good experimental
Technical Considerations for Additive Manufactured practices they used, as well as the clarity in which these
Medical Devices, further emphasizes the importance practices were explained (Summarized in Table 4).
of justifying statistical methodologies when validating
41
process control and demonstrating equivalence to Vilanova et al. used DOE for the classical problem of
predicate devices. Together, these frameworks establish a finding the right parameter combinations that maximize
40
comprehensive basis for experimental verification of safety a response variable. It highlights and utilizes the general
and performance in AM-produced orthopedic implants. guidelines for DOE (replication, randomization, etc.),
Therefore, 3D printing in medicine—particularly in it uses a design which is appropriate for the fitted model
orthopedics—serves as a valuable point of comparison for (CCD fitted to a second-order regression model), it has a
highlighting the lack of standardized experimental and reasonably high sample size, and the response function is
statistical practices in PBF-LB/M. well defined. Specifically, they used this design to solve a
multi-objective problem where the response variable is a
To compare the two fields, we sampled an additional function of the desirability of several process outputs. They
20 papers from orthopedic research where AM was reported regression results, the relevant statistics, and the
used to build various anatomical structures with the end design matrix along with the raw results of the experiment.
goal of patient implantation (which we shall refer to as Overall, they used the highest proportion of proper
“orthopedic engineering”). As we did before, we created a experimental techniques of those papers we have checked.
list of candidate papers for AM in Scopus. The search term 18
we used was “Orthopedics” AND “3D Printing.” A random The goal of the work of Zhang et al. is to use a
set of these papers was quickly assessed for relevance. Half combination of simulations and experimental designs to
of these papers were from 2016, and the rest were drawn find regions of the process parameter space which result
in stable prints. They developed printability maps via
from 2024, so that we could compare temporal trends theoretical simulations, which showed the combinations
as well as fields. We also analyzed these papers as we did of parameters that result in different types of defects
before, noting what types of experimental designs were (if any). They then chose points within each of these
used in each study, the type of analysis, etc. We compare regions to determine whether they achieve the predicted,
the proportion of papers using each design to PBF in the categorical result. The authors used a Latin hypercube
same years.
design and reported the design matrix along with the raw
We find that nearly all studies in orthopedic engineering results. Figure 2 shows an example of how to illustrate an
perform either single-factor experiments or full factorial experimental method.

Volume 1 Issue 4 (2025) 10 doi: 10.36922/ESAM025340021