Blog
About

Artificial Neural Networks in Mechanical Engineering by Miguel Abambres

data + data-driven analytical models in Mechanical Engineering
Unlocking Research Collabs

Revolution in Science: youtu.be/N6Vz78d9t2U

2
recommends
+1 Recommend
5
shares

Collection details

Feel free to suggest me new papers for the collection and to send me proposals for collaboration(s). Please check researchgate.net/project/Applied-Artificial-Intelligence

Editors

Editorials

  • Record: found
  • Abstract: found
  • Editorial: found
Is Open Access

👨🏽‍🏫 Free Courses / Seminars 🗣

 Miguel Abambres (corresponding) ,  Miguel Abambres (corresponding) ,  Miguel Abambres (corresponding) (2019)

Seeking opportunities to give (in portuguese, english or spanish) ' FREE ' * short courses / seminars at any higher education institution. As example, please find on ResearchGate (you need to be logged in) the programs of 2 possible topics (I´ve got many more I can send you upon request). I´d like to stay 1-4 weeks in a single city and teach/talk at any institution interested in my proposals. I aim to:

 

(i) change the way academic/scientific literature is published and assessed

(ii) teach (and promote the interest and motivation for) topics of my interest

(iii) get to know people interested in having scientific collaborations with me

 

I´d like to ask the interested ones (or who know institutions that might have any interest)to get in touch with me via ResearchGate and/or abambres@netcabo.pt, so that we can talk about the details and hopefully start arranging my short stay in your country.

 

Thanks so much!

__________________________________________________________________

depending on the distance from Lisbon, I might have a single requirement to give free seminar / courses: that the interested academic institution is able to financially support my stay or Lisbon - xxx - Lisbon flight.

    Bookmark
    • Record: found
    • Abstract: found
    • Editorial: found
    Is Open Access

    Scientific Collaboration

     Miguel Abambres (corresponding) ,  Miguel Abambres (corresponding) ,  Miguel Abambres (corresponding) (2019)

    Dear scientist, good morning!


    Would you like to co-author some eprints with me? 
    Since I am against peer-reviewed publishing (you can see why in this paper), I can only collaborate if we don´t publish our work in peer-reviewed journals (unless we are invited by the editor and the paper does not undergo peer-review).

    My role
    Developing Artificial Neural Network (ANN) models and write that part of the papers, for any possible nonlinear function (continuous or discrete/qualitative) you may wish to approximate. I will use a deterministic (non-stochastic) ANN-based software of my own, valid for interpolation-based approximations only (e.g., it does not handle time series forecasting). All research carried out with that tool can be read at Abambres' Lab.

    What are ANNs?  
    You can see this technique (belonging to Artificial Intelligence) as the most powerful regression model in the World, whose formulation is inspired by the way any brain processes information. Each designed ANN will give us a matrix-based formula to estimate the variable or function of interest.

    How to select a suitable problem?    
    You just have to imagine situations where the accurate determination of a variable of interest is quite costly and unfeasible in current practice (i.e., when employing the traditional experimental or numerical methods), meaning that the existing analytical models (if any) are too conservative or unsafe.


    Data Gathering
    Imagine you have a variable or function Y = Y (X1, X2, X3, ..., Xn), either qualitative/discrete or quantitative/numerical/continuous, that can be accurately determined (experimentally or numerically) as function of the independent variables X1, X2, X3, ..., Xn (you can mix quantitative and qualitative Xi variables, if needed).

    The most important result of our work
    My software will determine a formula that in less than a second is able to compute Y with high accuracy for any given (X1,..., Xn) provided as input.

    Minimum amount of data
    As in any regression technique, we need data  . You will have to send me at least max{2^n, 400} data points (X1,..., Xn; Y).

    Where to write the data?
    I´ll send you (along with instructions) a proper excel template for you to write your collected data.

     

    FINAL NOTE (IMPORTANT, though NOT FULLY COMPULSORY)
    --------------------------------------------------------
    In order to assure the generalization ability** 
    of the proposed ANN for virtually any case, would be important that the dataset you provide has the following features:

    (i) For at least 1 (the more the better) input variable, have ALL values DISTINCT WITHIN that variable(ie, as many as the no. of data points)
        
    (ii) Not include points too close (almost coincident) to each other in the corresponding hyperspace (compulsory)    

    **i.e. better accuracy for data points where at least 1 input variable take values inexistent (among all original values)
    in the original dataset you provide
    --------------------------------------------------------


    If you use MATLAB, this is the way to check the number of distinct values (ndv) in a vector of values (vv):

    vv = [define the vector of values];
    ndv = length(unique(vv));
    --------------------------------------------------------

      Bookmark

      Collection Information