Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Process Monitoring and Estimation for Selective Laser Melting
The quality of Selective Laser Melting (SLM) 3D printed parts depends on their thermal history. We aim to estimate this history for process monitoring and control purposes using a combination of model-based and data-driven estimation and machine learning techniques.
Keywords: Additive Manufacturing, 3D Printing, Estimation, Machine Learning, Model Order Reduction
Additive manufacturing (AM) processes have show great promise for creating complex, functional structures that are difficult or impossible to fabricate using conventional (subtractive) manufacturing techniques. SLM is a promising and widely used technique for 3D printing of metallic parts for use in e.g., aerospace, automotive, and medical domains. In SLM, a uniform layer of powder material is deposited on a substrate and a laser is used to melt and solidify the powder over a desired geometry. This process is repeated multiple times to build up a 3D part layer-by-layer.
The quality of the final printed parts depends strongly on the thermal history of the material during the build process. Excessive heat buildup during the printing process can cause thermal stresses, leading to warping and/or mechanical weakness in the final part, as well as excessive surface roughness. In many safety critical applications, e.g., aerospace engines, these process errors must be detected so the part can be flagged for additional inspection and potentially discarded. Alternatively, closed-loop control could be used to improve the quality of the part. However, only the temperature of the surface is measurable during the printing process, the temperature of the material below the surface must be estimated.
The flow of heat within the part can be modelled using the heat equation. However, the physics near the the meltpool, where the laser input liquefies the powered metal, is extremely difficult to model from first principles. Moreover, these models can be computationally expensive due to the need to track the entire temperature field. In this project, we will investigate (i) augmenting first principles models with data, e.g., using machine learning, to improve prediction accuracy and (ii) model order reduction techniques to reduce computational complexity.
If the final results are promising they can potentially be turned into a publication.
Additive manufacturing (AM) processes have show great promise for creating complex, functional structures that are difficult or impossible to fabricate using conventional (subtractive) manufacturing techniques. SLM is a promising and widely used technique for 3D printing of metallic parts for use in e.g., aerospace, automotive, and medical domains. In SLM, a uniform layer of powder material is deposited on a substrate and a laser is used to melt and solidify the powder over a desired geometry. This process is repeated multiple times to build up a 3D part layer-by-layer.
The quality of the final printed parts depends strongly on the thermal history of the material during the build process. Excessive heat buildup during the printing process can cause thermal stresses, leading to warping and/or mechanical weakness in the final part, as well as excessive surface roughness. In many safety critical applications, e.g., aerospace engines, these process errors must be detected so the part can be flagged for additional inspection and potentially discarded. Alternatively, closed-loop control could be used to improve the quality of the part. However, only the temperature of the surface is measurable during the printing process, the temperature of the material below the surface must be estimated.
The flow of heat within the part can be modelled using the heat equation. However, the physics near the the meltpool, where the laser input liquefies the powered metal, is extremely difficult to model from first principles. Moreover, these models can be computationally expensive due to the need to track the entire temperature field. In this project, we will investigate (i) augmenting first principles models with data, e.g., using machine learning, to improve prediction accuracy and (ii) model order reduction techniques to reduce computational complexity.
If the final results are promising they can potentially be turned into a publication.
In this project, our aim is to develop a process monitoring framework for SLM. We have an SLM machine instrumented with a pyrometer, CMOS camera, and thermal camera, and will build on previous projects that established first principles modelling frameworks for the temperature evolution during an SLM build process.
The objective are as follows:
- Learn about selective laser melting processes and our experimental platform;
- Become familiar with our existing framework for multi-layer modelling of SLM processes;
- Develop a model-based estimator for recovering the 3D temperature field in a part based on surface emissivity (thermal camera) measurements;
- Apply model order reduction techniques to the developed estimator to enable use in control applications;
- Investigate data-driven methodologies (e.g., machine learning) for improving the accuracy of the models, boosting estimation performance, or further reducing the computational complexity of the estimator.
The relative emphasis of goals 4 and 5 can be tailored based on the expertise and interests of the student.
In this project, our aim is to develop a process monitoring framework for SLM. We have an SLM machine instrumented with a pyrometer, CMOS camera, and thermal camera, and will build on previous projects that established first principles modelling frameworks for the temperature evolution during an SLM build process.
The objective are as follows:
- Learn about selective laser melting processes and our experimental platform;
- Become familiar with our existing framework for multi-layer modelling of SLM processes;
- Develop a model-based estimator for recovering the 3D temperature field in a part based on surface emissivity (thermal camera) measurements;
- Apply model order reduction techniques to the developed estimator to enable use in control applications;
- Investigate data-driven methodologies (e.g., machine learning) for improving the accuracy of the models, boosting estimation performance, or further reducing the computational complexity of the estimator.
The relative emphasis of goals 4 and 5 can be tailored based on the expertise and interests of the student.
We are looking for a talented and highly motivated student with an interest in control of complex processes, system identification, and machine learning.
- Knowledge of control/estimation and/or machine learning;
- Enrollment in a Masters program;
- Proficiency in English.
- Some knowledge of partial differential equations and model order reduction would be helpful but not necessary;
- No specific experience with manufacturing systems is necessary;
Please send your resume/CV (including lists of relevant publications/projects) and transcript of records in PDF format via email to dliaomc@ethz.ch, ebalta@ethz.ch, tihanyid@ethz.ch, rupenyan@inspire.ethz.ch, and bkavas@ethz.ch.
We are looking for a talented and highly motivated student with an interest in control of complex processes, system identification, and machine learning.
- Knowledge of control/estimation and/or machine learning;
- Enrollment in a Masters program; - Proficiency in English.
- Some knowledge of partial differential equations and model order reduction would be helpful but not necessary;
- No specific experience with manufacturing systems is necessary;
Please send your resume/CV (including lists of relevant publications/projects) and transcript of records in PDF format via email to dliaomc@ethz.ch, ebalta@ethz.ch, tihanyid@ethz.ch, rupenyan@inspire.ethz.ch, and bkavas@ethz.ch.