Hi, I lead a large program management in a manufacturing setting. Have recently gone through Julia and I see a big potential in making Julia work for Manufacturing Problem statements. What intermediate/Advanced projects should I undertake to build competency in Julia. I have already completed the tutorial.
4 Likes
Can you explain a little bit more about what that is about? Are these optimization problems? Or can you share a link that explains these kinds of problems?
1 Like
I am not able to attach any URLs in the post. So, I 'll try to describe few generalized problems most manufacturing factories and the factory personnel will always face.
These are known devils across most industry sectors.
- Golden Batch (defect free) product prediction and postmortem analysis of Non-golden Batch Product
- Engines/Models to recommend ideal process parameters to run factory machines
- Predicting When factory machine/equipment will fail or start to degrade based on Sensors/data acquisition modules monitoring equipment health.
Most of it do not always require a ML approach. Stats combined with control charts and lean principles give out good results initially. But they reach a bottle neck in the following scenarios to name a few:
- The products the factory manufactures require compliances which causes one to work with very strict guidelines on product specifications. Many cases optimization models fail to factor non-linear relations, and many give a non-feasible solution
- If the manufacturing process is so complex or has a lot of input features getting captured in which case conventional statistic models cannot describe the said complexity.
- Performance Inconsistency caused because input features are non-linear or are captured in different timestamps or the data may be measured manually or captured with human bias.
Factories will try to solve this with either Simulation or Optimization models. ML is combined with simulation in few cases or if simulation modelling is expensive in time and capital, then ML is preferred.
One exception is the use cases where we need to predict machine/equipment failure. In such a case, Optimization is not used as ML has seen a better outcome and adoption across the industry.
However please note, that most manufacturing companies will never post their problem statements online & explicitly mention their issues as there are various data confidentiality clauses. You can always try to connect with folks on Linkedin who work in factories and see if they are willing to share some high-level ideas.
Hope this helps.
You can also google - “ml in manufacturing”, there are many articles which can give you ideas as to what factories deal with.
4 Likes
I mean, look at Get started with Julia
There are a few free courses, e.g. Introduction to Machine Learning | JuliaAcademy and Deep Learning with Flux.jl | JuliaAcademy
and Decision Making Under Uncertainty with POMDPs.jl | JuliaAcademy .
Or you could look at: Julia on Exercism
Of course, statistics would be another field to look at.
Try to think of a reasonably small example that you want to solve using Julia.
3 Likes
Perhaps I am saying something banal, but as you are already doing this stuff, my suggestion is just start.. with your own project.
Start porting to Julia what looks the simplest cases, so you can learn by doing, and get a clever idea for the more complex cases…
11 Likes
I second what @sylvaticus said and like to add: Feel free to come back to this forum whenever you run into trouble, don’t know how to approach a problem, generally want to discuss your code and get feedback, etc. We are a friendly, helpful bunch 
Before posting, I’d recommend reading this to get the best experience here:
1 Like
Welcome to the community @AkshayJ. Great to see more industry people here.
Take a look at ModelingToolkit: GitHub - SciML/ModelingToolkit.jl: An acausal modeling framework for automatically parallelized scientific machine learning (SciML) in Julia. A computer algebra system for integrated symbolics for physics-informed machine learning and automated transformations of differential equations.
This is s huge framework and may already have something relevant for your industry.
Be aware of the relation to Dyad: [ANN] Dyad: A New Language to Make Hardware Engineering as Fast as Software.
See also this thread on recent development: ModelingToolkit V11 Library Split and Licensing, Community Feedback Requested. It may look like uncertainty, but to me it sounds promising.
Hi, Will definitely look at POMDPs tutorial. I do have tracks in progress in exercism. Thank you for suggesting.
As I was writing the post on describing Manufacturing problem statement, I realized the same thing.
However, with the constraints we have on which tech stack we can use and how the projects need to be maintained, I cannot directly use Julia on the current use cases in our factory.
I have started searching for Open-source datasets which closely reflect the issues I have highlighted in Manufacturing. My aim to use Julia on these Open-Source datasets and propose to my broader team to view Julia as an alternative to our current tech stack.
Another problem statement, I did realize where Julia may excel really well would be Material, Alloy and Product design. As these fields initially really heavy on Simulation and Optimization, Julia looks to be a very ideal fit.
Thank you @sylvaticus , @abraemer for your suggestions.
Will definitely look at SciML first. Dyad probably looks overwhelming for me. Will touch Dyad once I am able to handle SciML, JuMP and Flux at least. Thank you for the suggestion.