Need Recommendation for Machine Learning in Additive Manufacturing

Introductions
1

Hi everyone,

My name is Ali Vali, and I am an assistant instructor at British Columbia Institute of Technology in Vancouver, Canada. I am working in the mechanical engineering department. I am getting my Master’s degree in additive manufacturing at Penn State. I am interested in using machine learning in additive manufacturing for in-situ process monitoring. Would you please make some recommendations on which courses I should take?

2 Likes
2

I’d start by studying anomaly detection, that matches well with process monitoring.

It is taught as part of the Machine Learning Specialization (three courses), offered through Coursera.

You’ll need at least the rudiments of python programming skills as well, and a bit of statistics (mean value and standard deviations).

3

Hi Ali, Hope you are doing good.

I’m a senior AI engineer who has 10 + years of experience in AI/ML development.

I would be happy to help you to understand the AI better and faster.

Thanks,

Luis

4

Hi Lusi,

I am an assistant instructor at the British Columbia Institute of Technology in Vancouver, Canada. I am also studying at Penn State to get my master’s degree in additive manufacturing.

I am interested in in-situ process monitoring for additive manufacturing. Would you please help choose a machine learning course(s), a prompt engineering course (do I need this course?), and a Python course?

Thanks,

5

I hope you describe the problems you like to solve in more detail. Then, I can help you choose suitable courses.

And you need to learn at least Python, ML and DL. So I recommend these 3 courses.

https://www.coursera.org/learn/python.

https://www.coursera.org/specializations/machine-learning-introduction.

https://www.coursera.org/specializations/deep-learning.

6

Dear Luis,

Thank you again for your prompt response. I appreciate your time and support.

As I mentioned, I am currently an Assistant Instructor in the Mechanical Engineering Department at the British Columbia Institute of Technology (BCIT), and I am pursuing a Master’s in Additive Manufacturing with a focus on LPBF and EBPBF technologies.

I’m working with an industry partner who is interested in implementing in-situ process monitoring for their metal additive manufacturing systems. Specifically, we aim to:

  • Monitor and control laser/EB speed in real time
  • Track and regulate melt pool temperature
  • Capture images of each printed layer and compare them with the corresponding CAD model layers

Predict and detect potential delamination or part defects as early as possible.

My primary learning goal is to understand how to effectively integrate these sensing techniques, particularly thermal monitoring and layer-wise image comparison, into a closed-loop control system. I want to learn how to extract meaningful data (e.g., melt pool temperature, geometric deviations, spatter behaviour, power source speed, and adjust power density) from real-time sensors and use this data to dynamically adjust process parameters, such as power density and scan speed. I want to learn how to utilize machine learning to achieve this.

If you know of any specific courses, workshops, or training resources related to machine learning, sensor integration, image analysis, or thermal modelling that could help me develop these skills, I’d be grateful for your recommendations.

I’m also very open to suggestions or ideas I may not have considered. Your guidance is greatly appreciated as I dig deeper into this area.

Thank you again for your support, and I look forward to hearing from you.

Warm regards,

7

To me this does not appear to be a Machine Learning application. It’s a system control application.

Same concern. This is a control system task.

Machine Learning uses a large collection of labeled data, and learns a mathematical model for the system. This allows you to make predictions.

Perhaps the task you’re working on isn’t really well-suited to machine learning tools.

There are aspects of your project that could use machine learning methods (for example, using image analysis to identify anomalies and classify whether they are defects).

8

Hi Tom,

Thank you for your email. I understand all the points you mentioned. The reason I sent all the information is that Luis told me he could help more if he knew what I am trying to do.

I am trying to learn machine learning (ML) for in-situ process monitoring in additive manufacturing (AM). Our mechatronics department will handle all sensors and data collection. As an additive manufacturing professional, I must be able to analyze the collected data; that is my job, since the mechatronics team does not understand the science behind it.

This is what I am trying to do: collect data, for example, laser speed, teach the software what the optimal speed is, and ask it to adjust accordingly.

I am taking Dr. Andrew Ng’s course on YouTube (CS229), which talks about the mathematics of machine learning. This is a perfect refresher course for me. Now, what I am trying to learn is how to implement them in my application. These are the questions in mind:

  • What coding software should I learn? I notice that most people are using Python.
  • What level of coding proficiency is required to get started with ML for in-situ monitoring?
  • Besides Python, are there specific software platforms or frameworks I should learn for image and signal processing?
  • Should I be using simulation tools (e.g., ANSYS) alongside ML tools to complement experimental data?
  • What types of ML models are typically used for in-situ monitoring tasks?
  • How do I choose between supervised, unsupervised, and semi-supervised models for tasks like defect detection or melt-pool prediction?
  • Do I need to learn neural networks or deep learning from the start, or should I build a foundation with simpler models?
  • What is the difference between Machine Learning, Deep Learning, and Neural Networks in the context of in-situ AM monitoring?
  • Are there courses or learning pathways that include hands-on, project-based work with real AM or sensor datasets?
  • What I’m trying to do is establish interdependencies. Do I need to learn a neural network?
  • Is there a recommended workflow or pipeline for processing in-situ AM data with ML, from data acquisition to model deployment?
  • Are there courses, certificates, or workshops that provide hands-on practice with real datasets—for example, importing sensor data, training a model, and applying it to predict part defects or process instability?

I know this is a big area, and I may not even be asking the right questions. I’m very open to suggestions, whether it’s about where to start, what to prioritize, or even what I may be missing entirely.

I want to know what I don’t know.

Thank you again for your time and support. I greatly appreciate it and look forward to learning more from your perspective.

Warm regards,

9

Sorry, cannot answer your list of questions at this time.

You might find this article on Ars Technica interesting:

10

Thank you, Tom.

11

I think, you are seeing the problem from the bottom. You need to see the problem from top and learn the skills like python, machine learning, neural network as needed.

I don’t know the courses, workshops, or training resources for your exact purpose.

I learned control engineering. Let me analyze your problems.

  • I believe there are sensors for measuring laser/EB speed and melt pool temperature and there are nothing AI can help here.
  • For the comparison between the prototype and real manufactured( layer-wise image comparison), we can apply Computer Vision here. I think, this is a challenge only AI can solve. I can help you with this.
  • For the thermal modeling, you can use Ansys. AI can help modeling but not much.
  • Now we can measure inputs (using sensors) and outputs (using AI) and we know the plant. Now we can control. We can use neural network to model the mapping between inputs and outputs. Or we can use neural network to get the optimal values for the control inputs like speed and temperature.
  • We can apply AI for the control algorithm even though I think PID is enough.

I can help you with these scopes and other AI related problems as well.

Let me know what specific area you need my assistance.

Best,

Luis

12

That would be great, thank you Luis.