Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Evaluation of ML potential for modelling and process control of a reduction annealing process

Gimbringer, Vidar LU and Ziebeil, Björn (2026) In Master's Thesis in Mathematical Sciences FMAM05 20252
Mathematics (Faculty of Engineering)
Abstract
While many industrial processes have been thoroughly exposed to machine learning
models, the realm of powder metallurgy is still underexplored. This thesis aims to
investigate the potential for a machine learning model to predict the characteristics
of metal powders after a reduction annealing process using time series process
data from the furnace. A supervised learning model was developed to predict
six target variables: three particle size distributions, two chemical composition
properties, and a density. The pipeline involved filtering and preprocessing data,
model training, and performance evaluation.
Multiple families of machine learning methods were explored and implemented,
including linear methods, tree-based ensemble... (More)
While many industrial processes have been thoroughly exposed to machine learning
models, the realm of powder metallurgy is still underexplored. This thesis aims to
investigate the potential for a machine learning model to predict the characteristics
of metal powders after a reduction annealing process using time series process
data from the furnace. A supervised learning model was developed to predict
six target variables: three particle size distributions, two chemical composition
properties, and a density. The pipeline involved filtering and preprocessing data,
model training, and performance evaluation.
Multiple families of machine learning methods were explored and implemented,
including linear methods, tree-based ensemble methods, and sequence models.
Hyperparameter values were systematically tuned from a range of values to find
the best configuration; each model’s performance was then assessed using cross-
validated error metrics. Using SHAP analysis, it was possible to identify the most
prevalent and important features for predicting each output for each model.
Results show that performance varies for each target variable, and while no
model was completely dominant, the most successful predictions were made by
LSTM, XGboost, and the Elastic Net, with all models presenting some overlapping
features with the highest SHAP scores. The findings show that all three types of
models tested were applicable and offer promising potential for the further rollout
of machine learning in this field. (Less)
Popular Abstract
Can machine learning be used to predict industrial processes in real time?
Metallurgy is the study of metals and their technical applications, a field that
combines chemistry, physics, and engineering. The creation of metal pow-
ders is a complex and intensive process, and customers demand a high-quality
product with consistent properties. To make sure the product reaches stan-
dards, laboratory analysis is carried out regularly on finished powder samples
at factories to constantly monitor the quality of powder. However, this means
that any necessary adjustments can only be made after the process, when it
is already too late.
Imagine instead being able to estimate the product quality during the
process, allowing changes to be... (More)
Can machine learning be used to predict industrial processes in real time?
Metallurgy is the study of metals and their technical applications, a field that
combines chemistry, physics, and engineering. The creation of metal pow-
ders is a complex and intensive process, and customers demand a high-quality
product with consistent properties. To make sure the product reaches stan-
dards, laboratory analysis is carried out regularly on finished powder samples
at factories to constantly monitor the quality of powder. However, this means
that any necessary adjustments can only be made after the process, when it
is already too late.
Imagine instead being able to estimate the product quality during the
process, allowing changes to be implemented immediately. This could save
time, manpower, resources and cut down material waste. This thesis inves-
tigates whether machine learning can bridge this gap by predicting product
properties during the manufacturing process.
Using existing time series data from industrial furnaces, where variables
such as temperature, pressure and fan speed are measured and the corre-
sponding laboratory property measurements, predictive models were trained
and developed. Several groups of machine learning methods were used, each
approaching the problem from a different perspective.
After extensive training and evaluation, three models stood out with
particularly strong performances. These models were assessed and com-
pared using multiple error measurements and visualizations, showcasing their
strengths and weaknesses. To get an intuitive understanding of how each
model makes a prediction, interpretability methods were used to identify
which features had the greatest influence on the results
While no model could be said to be conclusively superior, the results show
that the models are capable of identifying patterns in the furnace data and
that reasonably accurate predictions of the powders properties are possible.
This work is an important first step towards real-time industrial predictions of metallurgy and indicates promising potential for further modelling and
implementation of machine learning in this field. (Less)
Please use this url to cite or link to this publication:
author
Gimbringer, Vidar LU and Ziebeil, Björn
supervisor
organization
course
FMAM05 20252
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Machine Learning, Supervised Learning, Regression, Metallurgy, Metal Powder, Reduction Annealing, Elastic Net, XGBoost, LSTM
publication/series
Master's Thesis in Mathematical Sciences
report number
LUTFMA-3606-2026
ISSN
1404-6342
other publication id
2026:E6
language
English
id
9221269
date added to LUP
2026-02-11 13:18:37
date last changed
2026-02-11 13:18:37
@misc{9221269,
  abstract     = {{While many industrial processes have been thoroughly exposed to machine learning
models, the realm of powder metallurgy is still underexplored. This thesis aims to
investigate the potential for a machine learning model to predict the characteristics
of metal powders after a reduction annealing process using time series process
data from the furnace. A supervised learning model was developed to predict
six target variables: three particle size distributions, two chemical composition
properties, and a density. The pipeline involved filtering and preprocessing data,
model training, and performance evaluation.
Multiple families of machine learning methods were explored and implemented,
including linear methods, tree-based ensemble methods, and sequence models.
Hyperparameter values were systematically tuned from a range of values to find
the best configuration; each model’s performance was then assessed using cross-
validated error metrics. Using SHAP analysis, it was possible to identify the most
prevalent and important features for predicting each output for each model.
Results show that performance varies for each target variable, and while no
model was completely dominant, the most successful predictions were made by
LSTM, XGboost, and the Elastic Net, with all models presenting some overlapping
features with the highest SHAP scores. The findings show that all three types of
models tested were applicable and offer promising potential for the further rollout
of machine learning in this field.}},
  author       = {{Gimbringer, Vidar and Ziebeil, Björn}},
  issn         = {{1404-6342}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master's Thesis in Mathematical Sciences}},
  title        = {{Evaluation of ML potential for modelling and process control of a reduction annealing process}},
  year         = {{2026}},
}