Hi everybody, as you know I am Electronics engineer with a strong background in control systems, instrumentation and industrial automation and also I am finishing my masters degree in mechanical engineering, you know I love the mechatronics engineering and their role in the actual technical world,
today I want to show you some of the results that I found in my masters thesis project called, "Design, implementation and evaluation of a expert systems applied for failure analysis in mechanical elements" (Axis, gears, bearings and boilers), in that thesis project I´m comparing three different inference engines as a artificial intelligence strategy for diagnosis and interpretation:
The finish document is here http://www.bdigital.unal.edu.co/45143/
And here there is a paper (you can download for free the pdf for the next 50 days) about this work published in the engineering failure analysis journal Science direct.
http://www.sciencedirect.com/science/article/pii/S1350630715001089
The finish document is here http://www.bdigital.unal.edu.co/45143/
And here there is a paper (you can download for free the pdf for the next 50 days) about this work published in the engineering failure analysis journal Science direct.
http://www.sciencedirect.com/science/article/pii/S1350630715001089
1. Clasic inference engine: based in modus tollens and modus ponens rules.
2. Fuzzy inference engine: Based on a previous comparation between sugeno and mamdani inference engines.
3. Bayesian inference engine, based on bayesian networks.
The system that I developed for axis has got 4 modules:Fracture, corrosion, wear and plastic modules.
Acordding to the different test that I did, I am very surprised because the bayesian inference engine have been the better behavior compares with the clasic and fuzzy inference engine, is amazing how the system solves complex cases of interpretation of evidence in many failure analysis process, Im using a lot of failure analysis cases from the National University of Colombia (AFIS group have been resolved cases for many industries like Oil & gas, automotive and complex heavy industries).
At the end of my thesis ,probably 2 months, :)I will post the final results, this research project also could be implemented in other engineering fileds (instrumentation, automation and control systems) with amazing results.
I think that in the near future the applied cibernetics as artificial intelligence will be the win ticket for many companies thart spend a lot of money and time designing experts systems for their engineering optimization, honestly I think that artificial intelligence is the future for those industries.
Do you want to see a print screen of my system?
1. An example of bayesian network just for fracture module in axis:
2. Take a look around of the wear module, few questions are necesary to identify the failure mode, based in the knowledge base:
3. After of the analysis according to the evidence for example for adhesive failure wear mode the system shows the following fault three analysis FTA.(for more information of FTA analsysis read the certified realiability engineer handbook)
The GUI was developed using Matlab/Simulink GUIDE (Bayesian and fuzzy) and C#+Amzi! Prolog (clasic inference). The work has been hard.
The most interesting result in this research masters project is the integration of a lot of my passions in one project as Electronics engineering, mechanical engineering, software develop and control systems, maintenance and reliability knowledge...
Some Oil & Gas companies are interested to implement this system in those industries for interpretation and diagnosis in their specific process (alarm management, Safety Instrumentation systems, control systems), I will be waiting for new challenges :)
