ISO/ASTM TR 52958:2026

• Abstract:

  This document provides a workflow comprising experimental procedures and flaw detection algorithms aimed at locating flaws in parts produced during the powder bed fusion-laser-based (PBF-LB) process of metals. It emphasizes the use of coaxial photodiode-based in-situ monitoring and statistical and clustering machine learning algorithms, particularly for detecting lack of fusion-induced flaws. The workflow delineates setting thresholds for statistical detection and determining the number of clusters for machine learning algorithms, utilizing intentional seeded flaws in parts. Validation procedures are provided through computed tomography scanner data. Hardware limitations and considerations for multi-laser processes are addressed, with attention to potential issues.

• Status: Published
• Publication Date: 21-May-2026
• ICS: 25.030 - Additive manufacturing
• Technical Committee: ISO/TC 261 - Additive manufacturing
• Drafting Committee: ISO/TC 261 - Additive manufacturing
• Current Stage: 6060 - International Standard published
• Start Date: 22-May-2026
• Completion Date: 22-May-2026

• Consolidates: CEN ISO/ASTM TR 52958:2026 - Additive manufacturing of metals - Powder bed fusion (PBF) - In-situ coaxial photodiode monitoring for lack of fusion flaw detection in PBF-LB (ISO/ASTM TR 52958:2026)
• Effective Date: 12-Feb-2026
• Consolidates: ISO 24070-2:2021 - Building and civil engineering sealants — Determination of cured thickness of one-component sealant — Part 2: Cylindrical cup test method
• Effective Date: 18-Mar-2023

ISO/ASTM TR 52958:2026 is an ISO/ASTM Technical Report on additive manufacturing of metals, published by ISO with ASTM International. It specifies a workflow for using coaxial photodiode in-situ monitoring in powder bed fusion-laser-based (PBF-LB) metal processing to detect lack of fusion flaws. The document is written for people who need to set up, tune, and validate a flaw-detection method for metal parts built by PBF-LB.

What does ISO/ASTM TR 52958:2026 specify?

ISO/ASTM TR 52958:2026 specifies a workflow for locating flaws in metal parts made by PBF-LB using coaxial photodiode signals, statistical detection methods, machine-learning clustering methods, and computed tomography (CT) validation. The document is organized around building calibration coupons, collecting and correcting sensor data, selecting detection thresholds or clusters, and checking results against CT data.

The main structure is:

• Clause 1 - Scope
• Clause 2 - Normative references
• Clause 3 - Terms and definitions
• Clause 4 - Significance and use
• Clause 5 - Design of coupons
• Clause 6 - Sensor description
• Clause 7 - Flaw detection algorithms
• Clause 8 - Implementation of customized algorithms to components with randomized flaws and the associated voxelization thereof
• Clause 9 - Case study: procedural implementation and validation results

Clause 5 separates intentionally seeded flaws from randomized or stochastic flaws. Clause 7 presents the statistical and clustering approaches used to interpret the photodiode signal. Clause 8 shows how the tuned workflow is applied to production-like parts and checked against CT through voxelization and a matching matrix.

What are the key requirements of ISO/ASTM TR 52958:2026?

ISO/ASTM TR 52958:2026 gives a step-by-step workflow rather than a single test method. The practical message is that flaw detection in PBF-LB should be developed on seeded coupons first, then transferred to randomized flaws, and finally checked against CT.

Build and use coupon sets for calibration

Clause 5 recommends two coupon sets: one with intentionally seeded flaws and one with randomized or stochastic flaws. Reference datum features such as grooves or notches can be added to help align CT data with the build geometry. In practice, this means the flaw location in the scan can be registered back to the same place in the CAD model.

Clause 5.2 describes intentionally seeded flaws as spherical or cylindrical voids introduced by CAD or process manipulation. The document uses three feature-size classes for these flaws: the minimum possible size for the process, plus 50 µm, and plus 100 µm. That matters because the seeded flaws are used to tune detection settings before the same settings are applied to real process defects.

Create randomized flaws by lowering energy density

Clause 5.3 says randomized or stochastic lack of fusion flaws can be produced by changing process parameters. The listed routes are reducing laser power, increasing hatching distance, increasing scanning speed, or increasing layer thickness. The document recommends aiming for a small relative density reduction of around 0.5% compared with nominal parts, so the calibration parts behave more like flawed production parts.

Collect the right sensor data

Clause 6 identifies a coaxial photodiode arrangement aligned with the laser beam path through a beam splitter. The sampling frequency is equal to or more than 60 kHz, and the recorded data also includes laser modulation and XY scanner position. In practice, the sensor must collect enough temporal detail to let the algorithms see signal perturbations, and the data must be corrected for chromatic and monochromatic distortion.

Use two statistical detection methods

Clause 7.2 presents two statistical approaches: absolute limits (AL) and short term fluctuations (STF). Both use a moving average and threshold bands to identify signal perturbations, which the document links to potential flaw locations. For a reader, this means the main tuning tasks are choosing the window length and setting upper and lower thresholds that separate normal melt-pool behavior from abnormal signals.

Use clustering algorithms as a second route

Clause 7.3 describes unsupervised clustering methods, specifically self-organizing map (SOM) and K-means. The document says the cluster count should be optimized before use, and it gives screening ranges for SOM and an example of K-means tuning. In practice, this lets a user compare a threshold-based method with a machine-learning grouping method and choose the one that best matches the seeded flaw pattern.

Validate with CT, voxelization, and a confusion matrix

Clause 8 uses CT data to validate the tuned algorithm on parts with randomized flaws. The process stacks layers into a volume, voxelizes that volume, labels voxels from indicator counts, and compares the result with CT using a matching matrix and a confusion matrix. For a quality or inspection workflow, this creates a traceable way to measure true positives, false positives, false negatives, and true negatives.

What terms does ISO/ASTM TR 52958:2026 define?

ISO/ASTM TR 52958:2026 defines several terms that are central to the workflow:

• clustering algorithm - an unsupervised machine-learning method that groups unlabelled input data by similarity.
• coaxial photodiode arrangement - a sensor arrangement on a PBF-LB machine aligned with the laser beam path.
• computed tomography (CT) - a non-destructive examination method that reconstructs a 3D volume data set or 2D cross-sections from radiographic projections.
• flaw indicator - an indicator corresponding to the predicted location of a flaw.
• lack of fusion - process-induced porosity with not fully melted or fused powder particles on the previously deposited substrate.
• intentionally seeded flaw - a flaw deliberately created through CAD, processing-parameter manipulation, or insertion of an artificial object.
• data registration - the process of aligning data and assigning a persistent identification to the aligned data set.

Who uses ISO/ASTM TR 52958:2026?

ISO/ASTM TR 52958:2026 is used by additive manufacturing engineers, process developers, and machine operators working with metal PBF-LB systems. It is also relevant to quality managers, inspection teams, and researchers who need to validate in-situ monitoring against CT.

In practice, these users apply the document when they:

• design seeded and stochastic test coupons
• collect coaxial photodiode signals during builds
• tune threshold or clustering algorithms
• register CT data to the build geometry
• compare detected indicators with actual flaws
• assess whether a monitoring setup is suitable for a given machine or material

Which standards are used with ISO/ASTM TR 52958:2026?

ISO/ASTM 52900, Additive manufacturing - General principles - Fundamentals and vocabulary, is the normative reference named in the document. It supplies the general additive manufacturing vocabulary used alongside the terms in this Technical Report.

ISO/ASTM TR 52906, Additive manufacturing - Non-destructive testing - Intentionally seeding flaws in metallic parts, is cited for the creation of seeded flaws. It supports the coupon design work in Clause 5.2 by linking this report to a seeding approach already used for metallic parts.

What does the ISO/ASTM TR 52958:2026 document contain?

ISO/ASTM TR 52958:2026 contains a full workflow description, with figures showing the sensor layout, threshold-based signal interpretation, clustering-based interpretation, voxelization, and overall validation sequence. It also includes worked examples of coupon designs, signal plots, and CT comparison images used to calibrate the methods.

The document further contains:

• a case study on Hastelloy X parts built with seeded and randomized flaws
• example process settings used to create calibration parts
• figures for AL, STF, SOM, and K-means interpretation
• a voxel-based comparison method for CT validation
• a confusion matrix and accuracy calculation for assessing detection results

The case study shows how the workflow is applied in practice, including how seeded flaws are used to tune the algorithm before testing parts built with reduced energy density.

Technical
Report
ISO/ASTM TR 52958
First edition
Additive manufacturing of metals —
2026-05
Powder bed fusion (PBF) — In-situ
coaxial photodiode monitoring for
lack of fusion flaw detection in PBF-
LB
Fabrication additive de métaux — Fusion sur lit de poudre —
Surveillance par photodiode coaxiale in situ pour la détection de
défauts de fusion en PBF-LB
Reference number
© ISO/ASTM International 2026
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on
the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address below
or ISO’s member body in the country of the requester. In the United States, such requests should be sent to ASTM International.
ISO copyright office ASTM International
CP 401 • Ch. de Blandonnet 8 100 Barr Harbor Drive, PO Box C700
CH-1214 Vernier, Geneva West Conshohocken, PA 19428-2959, USA
Phone: [phone] Phone: [phone]
Fax: [phone]
Email: [e-mail] Email: [e-mail]
Website: www.iso.org Website: www.astm.org
Published in Switzerland
© ISO/ASTM International 2026 – All rights reserved
ii
Contents Page
Foreword .iv
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 1
4 Significance and use . 2
5 Design of coupons . 3
5.1 General .3
5.2 Intentionally seeded flaws .4
5.3 Randomized/stochastic flaws .6
6 Sensor description . 6
7 Flaw detection algorithms . 7
7.1 General .7
7.2 Statistical algorithms .8
7.2.1 General .8
7.2.2 Absolute limits (AL) .9
7.2.3 Short term fluctuations (STF) .11
7.2.4 Workflow .11
7.3 Machine-learning approach: clustering algorithm . 13
7.3.1 Self-organizing map (SOM) .14
7.3.2 K-means . 15
8 Implementation of customized algorithms to components with randomized flaws and
the associated voxelization thereof .16
9 Case study: procedural implementation and validation results . 19
Bibliography .25

© ISO/ASTM International 2026 – All rights reserved
iii
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards
bodies (ISO member bodies). The work of preparing International Standards is normally carried out through
ISO technical committees. Each member body interested in a subject for which a technical committee
has been established has the right to be represented on that committee. International organizations,
governmental and non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely
with the International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
The procedures used to develop this document and those intended for its further maintenance are described
in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for the different types
of ISO documents should be noted. This document was drafted in accordance with the editorial rules of the
ISO/IEC Directives, Part 2 (see www.iso.org/directives).
ISO draws attention to the possibility that the implementation of this document may involve the use of (a)
patent(s). ISO takes no position concerning the evidence, validity or applicability of any claimed patent
rights in respect thereof. As of the date of publication of this document, ISO had not received notice of (a)
patent(s) which may be required to implement this document. However, implementers are cautioned that
this may not represent the latest information, which may be obtained from the patent database available at
www.iso.org/patents. ISO shall not be held responsible for identifying any or all such patent rights.
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and expressions
related to conformity assessment, as well as information about ISO's adherence to the World Trade
Organization (WTO) principles in the Technical Barriers to Trade (TBT), see www.iso.org/iso/foreword.html.
This document was prepared by Technical Committee ISO/TC 261, Additive manufacturing, in cooperation
with ASTM Committee F42, Additive Manufacturing Technologies, on the basis of a partnership agreement
between ISO and ASTM International with the aim to create a common set of ISO/ASTM standards on
Additive Manufacturing, and in collaboration with the European Committee for Standardization (CEN)
Technical Committee CEN/TC 438, Additive manufacturing, in accordance with the Agreement on technical
cooperation between ISO and CEN (Vienna Agreement).
Any feedback or questions on this document should be directed to the user’s national standards body. A
complete listing of these bodies can be found at www.iso.org/members.html.

© ISO/ASTM International 2026 – All rights reserved
iv
Technical Report ISO/ASTM TR 52958:2026(en)
Additive manufacturing of metals — Powder bed fusion (PBF)
— In-situ coaxial photodiode monitoring for lack of fusion
flaw detection in PBF-LB
1 Scope
This document provides a workflow comprising experimental procedures and flaw detection algorithms
aimed at locating flaws in parts produced during the powder bed fusion-laser-based (PBF-LB) process of
metals. It emphasizes the use of coaxial photodiode-based in-situ monitoring and statistical and clustering
machine learning algorithms, particularly for detecting lack of fusion-induced flaws. The workflow
delineates setting thresholds for statistical detection and determining the number of clusters for machine
learning algorithms, utilizing intentional seeded flaws in parts. Validation procedures are provided through
computed tomography scanner data. Hardware limitations and considerations for multi-laser processes are
addressed, with attention to potential issues.
2 Normative references
The following documents are referred to in the text in such a way that some or all of their content constitutes
requirements of this document. For dated references, only the edition cited applies. For undated references,
the latest edition of the referenced document (including any amendments) applies.
ISO/ASTM 52900, Additive manufacturing — General principles — Fundamentals and vocabulary
3 Terms and definitions
For the purposes of this document, the terms and definitions given in ISO/ASTM 52900 and the following
apply:
ISO and IEC maintain terminology databases for use in standardization at the following addresses:
— ISO Online browsing platform: available at https:// www .iso .org/ obp
— IEC Electropedia: available at https:// www .electropedia .org/
3.1
clustering algorithm
unsupervised machine learning methods with unlabelled input data is grouped by similarity
3.2
coaxial photodiode arrangement
type of sensor arrangement on the powder bed fusion-laser-based machine aligned with the laser beam path
3.3
computed tomography
CT
non-destructive examination technique capturing radiographic projections of an object at various rotational
angles followed by mathematically reconstruction to produce a three-dimensional volume data set or one or
more two-dimensional cross-sectional images

© ISO/ASTM International 2026 – All rights reserved
3.4
data alignment
process of transforming different sets of geometrically or temporally related data into a single, global
coordinate system
3.5
data registration
procedure of data alignment and assignation of a persistent identification to the aligned data set
3.6
ex-situ analysis
measurement procedure performed after the completion of the build cycle
3.7
flaw indicator
indicator corresponding to the location of flaws predicted by the flaw detection algorithm
3.8
lack of fusion
type of process-induced porosity with not fully melted or fused powder particles onto the previously
deposited substrate.
[SOURCE: ASTM E3166-20, 3.4.7]
3.9
reference datum feature
notch, groove, or similar feature added to the geometry in a seeded flaw coupon to ease data alignment and
registration of porosity locations for CT-scan ex-situ analysis
3.10
intentionally seeded flaw
act of intentionally creating flaws through computer-aided design or manipulation of designated processing
parameters, resulting in the placement of the anticipated flaw or the act of intentionally creating a flaw
through the insertion of an artificial object
4 Significance and use
A workflow for indirect flaw detection and analysis documentation during PBF-LB is provided by using
the signals received from a coaxial photodiode that can detect flaws, including lack of fusion, in fabricated
components.
These flaws may have detrimental effects on the mechanical performance of fabricated parts. The workflow
of this document provides a procedure to identify the range of upper and lower thresholds required for the
statistical detection algorithms to identify stochastic lack of fusion defects induced during the process. It
provides a procedure to identify the number of clusters required for machine learning detection algorithms.
In the validation procedure, the datasets collected from a CT scanner that are registered and voxelized
can be used. It is noted that the size of detectable flaw, as determined by the procedure outlined in this
document, is contingent upon the resolution and frequency of the hardware employed, specifically a co-
axial photodiode and its associated data acquisition card. For instance, when utilizing a commonly available
photodiode with a frequency of 60 kHz, the procedure and algorithms specified by this document are unable
to detect flaws smaller than 100 μm.
In general, an in situ photodiode installed coaxially provides information from the process signature and
flaws. However, the recorded in situ data needs to be corrected to remove chromatic and monochromatic
distortion. The corrected data analyse by two main algorithms to identify flaws:
a) statistically, and
b) by machine learning.
© ISO/ASTM International 2026 – All rights reserved
These algorithms can be systematically optimized and customized to detect lack of fusion flaws. To this
end, intentionally seeded flaws are first added to the computer-aided design (CAD) of coupons to tune
the parameters of the algorithms. Then, the customized algorithm is tested by detecting randomized/
stochastic flaws created by powder bed fusion-laser based with intentionally decreased energy density.
The comparison of detection results could be analysed by algorithms with the CT data applied through a
volumetric approach to identify the randomized/stochastic flaws. A flowchart illustrating the progression
in this document is shown in Figure 1.
Figure 1 — Schematic for calibration flow needed for detecting the lack of fusion flaw
5 Design of coupons
5.1 General
To customize and calibrate the detection algorithms systematically, two sets of coupons are suggested.
Reference datum features can also be added to the geometry to ease data alignment and data registration
of porosity locations in the ex situ analysis that is CT-scan in this practice. Figure 2 represents some
suggestions for registry notches/grooves.

© ISO/ASTM International 2026 – All rights reserved
a) Added vertical and horizontal registry grooves b) Added inclined notches
Figure 2 — Addition of registry grooves and inclined notches to the geometry of the coupon
5.2 Intentionally seeded flaws
The effect of the lack of fusion flaw can be mimicked by embedding intentional seeds/voids in the coupons.
[1]
According to ISO/ASTM TR 52Same as above906 , for creating these seeds, various sizes, distributions,
and geometries of seeds can be added to the computer-aided design. Two forms of spherical and cylindrical
intentional seeded flaws can be considered where the size of spherical flaws is identified by their diameter
and the size of cylindrical flaws is identified by their cross-sectional diameter and height. Note that the
minimum size of the intentional flaw is dictated by the PBF-LB restrictions. It is, however, recommended
that the feature size of flaws is set in the computer-aided design model to three different classes:
The minimum size possible to be made by the PBF-LB (for example 100 µm for the diameter of spheres and
100 µm for height and diameter of cylinders) depending on the resolution and laser spot size:
a) the minimum value plus 50 μm (for example 150 µm for the above-mentioned parameters);
b) the minimum value plus 100 µm (for example 200 µm for the above-mentioned parameters).
Note that within the layers in which the intentional flaws are made, an optimum down-skin parameter is
used to endure the mechanical integrity of the intentional flaw. The capping layer, however, can have no
down-skin setting.
Two examples of intentionally seeded flaws are shown in Figure 3. Figure 3 a) and b) represent two
dimensional cross sections of samples showing the distribution of the intentionally seeded flaws (cylindrical
and spherical), respectively. In Figure 3 a), six nominally identical sets of three sizes of cylindrical flaws (Ø,
H = 200 µm, Ø, H = 150 µm, and Ø, H = 100 µm are shown; in Figure 3 b), six nominally identical sets of three
sizes of spherical flaws (Ø = 200 µm, Ø = 150 µm, and Ø = 100 µm) where Ø is the diameter and H is the
height, in microns, are demonstrated; in Figure 3 c), the capping layer of spherical flaws are represented.

© ISO/ASTM International 2026 – All rights reserved
a)  Cylindrical type b)  Spherical type
c)  Schematic of spherical flaws showing the capping layer
Key
A sets (clustered intentional voids)
B capping layer
NOTE 1 All dimensions are in SI coordinate system.
NOTE 2 Figures 3 a) and 3 b) are published under an open access CC by 4.0 license.
Figure 3 — Two dimensional cross sections of samples showing the distribution of different types of
intentionally seeded flaws
© ISO/ASTM International 2026 – All rights reserved
5.3 Randomized/stochastic flaws
Randomized/stochastic flaws can be achieved normally because of process anomalies or by altering process
parameters in which the lack of fusion flaws are created by decreasing the energy density during the build
cycle. For creating randomized lack of fusion flaws, four scenarios are recommended:
a) reducing the laser power;
b) increasing the hatching distance;
c) increasing the scanning speed;
d) increasing the layer thickness.
These alterations depend on the material. It is, however, recommended that the produced parts with altered
process parameters exhibit a relative density reduction of around 0,5 % compared to parts built with
nominal process parameters.
6 Sensor description
The sensor embodied in this document is a coaxial photodiode arrangement with a sampling frequency
of equal or more than 60 kHz. The coaxial photodiode arrangement is aligned with the laser beam path
through a beam splitter (see Figure 4).
NOTE 1 Photodiodes can capture light intensity signals from the melt pool in different wavelengths; however, the
preferred wavelength range for most metallic alloys to capture melt pool light intensity is normally in the visible and
near-infrared ranges (between 750 nm to 900 nm).
In addition to the light intensity data, laser modulation and XY scanner position are recorded and stored in
the associated personal computer. Intensity and geometry calibrations of dataset are also required.
NOTE 2 Details of the data calibration routines are out of the scope of this practice.
NOTE 3 The intensity correction for the coaxial data can be implemented because of the chromatic aberration
phenomenon. It is initiated because the wavelength of light intensity recorded by the photodiode is not the same as the
wavelength optimized for the scanner mirror and f-theta lens or dynamic focusing units.
NOTE 4 The intensity and geometry corrections are dependent on the commercial system and are normally
implemented by the original equipment manufacturer’s monitoring system.

© ISO/ASTM International 2026 – All rights reserved
Key
1 laser
2 scanner mirror
3 F-Thena lense
4 roller (recoater)
5 powder
6 build plate
7 beam splitter
8 along the beam path
9 co-axial sensor
Figure 4 — Position of the coaxial sensor in the PBF-LB setup
7 Flaw detection algorithms
7.1 General
Two different approaches are recommended for detecting flaws: statistical algorithms and clustering
machine learning-clustering algorithms.
The statistical algorithm works based on the threshold method, and the clustering algorithm distributes
data into different groups. The workflow to detect the flaws was demonstrated in Figure 1 and is the
following:
Step 1: The preferred algorithm can be applied to the data collected during the printing of the samples with
intentionally seeded flaws.
© ISO/ASTM International 2026 – All rights reserved
Step 2: The detection result is compared with the design and CT scan data to customize/calibrate the
algorithm parameters such as moving average windows length and thresholds.
NOTE The setting of CT scan can normally be optimized in terms of resolution.
Step 3: The customized algorithm stemmed from Step 2 can be used for the detection of randomized flaws
in the fabricated samples.
Step 4: The detection results of Step 3 are validated by CT scan through the volumetric approach and
confusion matrix which is explained in Clause 9.
As an example, several statistical and machine-learning algorithms that are suitable for this practice are
discussed in 7.2 and 7.3. Additionally, the volumetric approach is used to compare the result with CT, which
is discussed in Clause 9.
7.2 Statistical algorithms
7.2.1 General
Statistical algorithms work based on the selection of threshold levels, and their central feature is the moving
[2]
average. Thus, it is critical to have an optimized workflow to identify a threshold range and the window
size/length of the moving average by end users. In Figure 5, the application of the threshold range to the
photodiode's signal is shown. As seen, some ripples exist outside the upper and lower thresholds where each
data sample is associated with a geometrical point in the XY cartesian coordinate. Any processed signal (for
example moving averaged) greater than the upper threshold and lower than the lower threshold is defined
and is called a signal perturbation in this document [Figure 5 b)]. Signal perturbations are associated with an
abnormality in the process and are normally corresponding to the location of flaws. Thus, the perturbation
can be mapped to the geometry of samples and visualized with a different colour [for example as a yellow
indicator shown in Figure 5 c)]. To map the perturbation with potential flaws in the fabricated components,
a proper detection algorithm is needed to be selected from absolute limits (AL) and short-term fluctuation
(STF). Two statistical algorithms are discussed in 7.2.2 and 7.2.3.
NOTE The capping layer of intentional seeded flaws [see Figure 5 c)] is used to optimize the value of upper and
lower thresholds and data window length since it represents higher signal perturbations.

© ISO/ASTM International 2026 – All rights reserved
Key
X1 time
X2 horizontal size of a layer
Y1 light intensity (a.u.)
Y2 vertical size of a layer
A photodiode time series data along with upper and lower thresholds
B magnified signal representing signal perturbation
C one layer of the sample including indicators
1 photodiode signal
2 m
...