Detection and classification of internal flaws in laser powder bed fusion: application of in-situ monitoring for quality control of Hastelloy X builds
Licentiate thesis, 2021

Additive manufacturing technologies, in particular laser powder bed fusion (LPBF), have received much attention in recent years due to their multiple advantages over traditional manufacturing. Yet, the usage of additively manufactured products is still quite limited, mainly due to two factors: the low repeatability, which is particularly relevant for applications where high performance is required from the materials, and the typically low productivity, particularly relevant for products with a substantial production volume.

The main factor that affects repeatability and compromises the performance of the materials is the presence of flaws. Hence, to assess the quality of a product and to predict its performance, it is crucial to recognize which flaws are present and ensure their detectability. Moreover, if the flaws can be detected during the manufacturing process, corrective actions can be taken. In this thesis, internal flaws were deliberately created in LPBF manufactured material to assess their detectability via in-situ monitoring. Two main routes of deliberate flaw formation have been identified while preserving flaw formation mechanisms; therefore, this thesis is split into two parts, according to the approach employed to create flaws.

Flaws are generated systematically if inadequate process parameters are employed. By varying the processing conditions, different types, amounts and sizes of flaws are created. By monitoring the manufacturing process with long-exposure near-infrared imaging and applying supervised machine learning, it was possible to distinguish process conditions that generate the different flaw categories with accuracy, precision and recall of at least 96%.

Flaws are created stochastically as a result of the redeposition of process by-products on the build area. It was found that substantial amounts of flaws can be provoked through this route when increasing the nominal layer thickness in the build, thus enabling the validation of the monitoring system in their detection. After applying an image analysis algorithm to all the images output from in-situ monitoring in three builds, it was possible to identify trends in the spatial distribution of spatter redeposits. Ex-situ inspection and material characterization provided cross-check for the distribution of flaws.

The low productivity of LPBF makes it less competitive in applications with moderate to high production volumes. This issue is briefly addressed in this thesis. Even though one of the main approaches to increase productivity is to tune the main process parameters, dissimilar strategies were identified in the literature towards this goal. Thus, parametrization of build rates was done and applied to the processing conditions deemed to provide material with acceptable quality, based on the quantity and types of flaws present. The material manufactured in these conditions was characterized, and it was found that substantially different microstructures can be achieved within the process window, depending on the build rate.

melt pool


flaw detection

nickel-based superalloy

powder bed fusion

lack of fusion



Additive manufacturing

process monitoring


Virtual Development Laboratory (VDL), Chalmers Tvärgata 4C
Opponent: Prof. Lars-Erik Rännar, Mid Sweden University


Claudia de Andrade Schwerz

Chalmers, Industrial and Materials Science, Materials and manufacture

Schwerz, C.; Nyborg, L.; "Identification of internal flaws in laser powder bed fusion by a neural network employed on near-infrared long-exposure images acquired by in-situ monitoring"

Schwerz, C.; Schulz, F.; Nyborg, L.; "Increasing productivity of laser powder bed fusion manufactured Hastelloy X through modification of process parameters"

Subject Categories

Materials Engineering

Manufacturing, Surface and Joining Technology

Metallurgy and Metallic Materials

Areas of Advance

Materials Science

IMS: 2021-12



Virtual Development Laboratory (VDL), Chalmers Tvärgata 4C


Opponent: Prof. Lars-Erik Rännar, Mid Sweden University

More information

Latest update