I might actually differ from the answer given above. These famous softwares like ABAQUS has got almost everything and moreover they update themselves and keep on incorporating new methods and theories regularly. But for marketting you have some of the issues with these softwares where you can work on. 

Firstly, most of the users of these softwares are students, especially in India and nearby countries license of the software is very costly for individuals, even the student version. and student version should be more versatile as students play around the code more than others.

Secondly, some of the recently famous theories are not updated in famous softwares, thats the reason you will find WARP3d more famous in research group than ABAQUS, although the difference will be too less. Because WARP3D lacks visualization, however just for example direct calculation of Weibull Stress which is pretty famous and important parameter in fracture, can only be done in WARP 3D, thanks to Dodds Jr for his coding,

Damage mechanics is another part which is day by day improving and not much is available in commercial softwares. SO you need to address these things and if you can give some controls to users your product will work more than commercials who lock the coding part. ABAQUS has a soln to it by coupling codes with subroutines, but those also have got their own difficulties.

There are many things which are missed by these softwares but I dont know where you are interested.

Then to your question that what type of knowledge is required is again related to my first part of answer i.e your marketing strategy. If you are going to market your product by  addressing for example damage mechanics and focus on that only because in versatility you CANNOT beat commercial packages, They have too many modules to solve almost all types of problems.

Moreover coding an FEM is not just solving F = KU, its much more than that, your code need to tell user that where he has given wrong values, why the soluton is not converging instead of just saying that solution is not converging. These things are again unbeatable but I wont say it cannot be done. Atleast for one particular field which might be your marketing point it is possible.

So when in a normal distribution the middle part is covered by commercial packages your chances are greater at the tails of the distribution if you know what i mean, and then by satisfying specific consumers you will grow with experience and experties.

One more problem of commercial packages I forget to tell is the analysis time This is a huge issue and a compromise you can again see in WARP 3D. WARP 3d has greater speed on the cost of visualization, This is a big issue as the time I am talking about can be days or even a month and then after waiting for that long you come to know that results are wrong.

Another point to focus on can be mixing two things for e.g. diffusion with mechanics, which is needed for hydrogen embrittlement, or non-local values of damage parameters. These are the things which is needed. at least if you give me these things I will buy your product and yes it has to be user friendly not just gibberish coding, :) .

All the best 

Before answering the questions you posed, Mohammad, two important questions are: "What would your software do better than currently available options?" and "Who would most benefit from what your software would do better?" When you have good answers to those questions then some of your other questions become easier to address.

For example, if your software would be tailored to one specific type of engineer (you mention mechanical engineers in your question) then it may not be as marketable as a more broadly applicable software. However, if you provide more intuitive user interfaces that would make the software more efficient than other available options to a mechanical engineer then that may be okay.

In the civil engineering world, one example of a FE software that has been developed over the years as a simpler but accurate alternative is VecTor by Prof Vecchio's research group at the University of Toronto (http://www.civ.utoronto.ca/vector/software.html). I don't see any reason why you would have to be university-based to develop your own software, however.

Best wishes!

I might actually differ from the answer given above. These famous softwares like ABAQUS has got almost everything and moreover they update themselves and keep on incorporating new methods and theories regularly. But for marketting you have some of the issues with these softwares where you can work on. 

Firstly, most of the users of these softwares are students, especially in India and nearby countries license of the software is very costly for individuals, even the student version. and student version should be more versatile as students play around the code more than others.

Secondly, some of the recently famous theories are not updated in famous softwares, thats the reason you will find WARP3d more famous in research group than ABAQUS, although the difference will be too less. Because WARP3D lacks visualization, however just for example direct calculation of Weibull Stress which is pretty famous and important parameter in fracture, can only be done in WARP 3D, thanks to Dodds Jr for his coding,

Damage mechanics is another part which is day by day improving and not much is available in commercial softwares. SO you need to address these things and if you can give some controls to users your product will work more than commercials who lock the coding part. ABAQUS has a soln to it by coupling codes with subroutines, but those also have got their own difficulties.

There are many things which are missed by these softwares but I dont know where you are interested.

Then to your question that what type of knowledge is required is again related to my first part of answer i.e your marketing strategy. If you are going to market your product by  addressing for example damage mechanics and focus on that only because in versatility you CANNOT beat commercial packages, They have too many modules to solve almost all types of problems.

Moreover coding an FEM is not just solving F = KU, its much more than that, your code need to tell user that where he has given wrong values, why the soluton is not converging instead of just saying that solution is not converging. These things are again unbeatable but I wont say it cannot be done. Atleast for one particular field which might be your marketing point it is possible.

So when in a normal distribution the middle part is covered by commercial packages your chances are greater at the tails of the distribution if you know what i mean, and then by satisfying specific consumers you will grow with experience and experties.

One more problem of commercial packages I forget to tell is the analysis time This is a huge issue and a compromise you can again see in WARP 3D. WARP 3d has greater speed on the cost of visualization, This is a big issue as the time I am talking about can be days or even a month and then after waiting for that long you come to know that results are wrong.

Another point to focus on can be mixing two things for e.g. diffusion with mechanics, which is needed for hydrogen embrittlement, or non-local values of damage parameters. These are the things which is needed. at least if you give me these things I will buy your product and yes it has to be user friendly not just gibberish coding, :) .

All the best 

Mohammad,

The commercial software you mentioned are definitively not a result of one-man job. Tremendous resources (dozens of developers + years of FTE) are required to build such codes. Academic simulators can be developed faster but are likely to have narrower functionalities with less robustness and efficiency.  My recommendation to you is to start with a simple PDE (maybe elliptic equation with simple BC’s in 1D). MATAB is a good programming software as it already has many useful built-in functions related to EFM from gridding to solvers.

Hope this helps.

Bruhl, your comment has two points: Customer needs and difference of the software with others. I agree with you but how?  as a user what do you need in a FEM software that does not exist in these softwares now? 

For example Abhishek said in ABAQUS, Weibull Stress is not embedded whereas it is very important. I agree with him, I always ask why important theories does not exist in it. Debugging , analysis time and mixing problems are the other points that Abhishek mentioned. 

Dear Hussein, 

You mentioned "academic simulator can be developed faster but are likely to have narrower functionalities with less robustness and efficiency", I agree with you about speeding but what is your reason about disadvantages? I think MATLAB can not be used as a programming software for big software like ABAQUS. In recent years, many programming languages are developed. For example, in Wikipedia some of them are listed. Do you think C#, FORTRAN and Python are best programming languages for FEM softwares? 

Hojjat, I read this conversation. I followed it but the question is not about my need. I want to know what are the requirements of developing an independent FEM software? however, What do you think about the future of open source codes? Is it possible one day, these softwares close their source?! what about their results?

Abhishek, I like to know what other things are missed in these softwares?

Regards

https://en.wikipedia.org/wiki/List_of_programming_languages

Mohammad, I can't answer the question about what customer needs you could uniquely address with a new software. As a civil engineer, I'm generally satisfied with the FE software I use most regularly (LS-DYNA) and would have to be convinced that it would be worth my time to learn a new software - that is, what benefits would I get for the time spent learning.

The area of needed improvement for my current research is actually material models within the FE software but LS-DYNA has been written in such a way that user defined material models are possible and its library of available material models is quite vast and continues to grow. Material models that require inputs that can be experimentally measured is my personal preference. Having material models that accurately account for damage is another area that is improving but still has more work needed to be done.

From your comments, I get the impression that you already have some ideas about what you would be able to do better with your software. Perhaps others can offer more ideas about what they'd like to see in a new, improved FE software (such as some of the comments already included here by others). 

Best wishes.

As said by Bruhl, developing a new package needs lots of focus and time and still one will lack the experience which commercials have. So as I said earlier that for initial stage a code for specific purpose can be built and when its a success you may expand. 

There are many local things, theories not available in commercials although which can be coded by subroutines to be used with commercials but these subroutines needs expertise in coding language and also needs focus to justify the code and then use it. This is the marketing point. If you provide a needed parameter ready to be used in a code. People will take it; for instance you can just write subroutines for ABAQUS and I am sure you will get customers for that.

Nevertheless, these still does not one problem of commercials, which is their speed to solve the problem. ABAQUS has come up with coalition with NVIDIA as quadro and Tesla which decreases required time to solve equations tremendously but these are not available to many people. 

Are you getting my point? providing a solver which has good speed on a regular machine can be achieved probably only if you compromise on visualization. and people dont mind that and then you provide them speedy package with inbuilt parameters to be extracted. sounds good to me. Another issue which comes up is that this will be a new package, perhaps very raw in user interface as you are not giving much of visualization. So, user will need to learn to operate the package. This is a big deal as users are reluctant to learn new software when they know a good one.  So it depends on what you provide to them and how you do that.

For e.g. simulating crack growth in ABAQUS is only by LEFM method or by cohesive elements or by writing a code i.e. VUMAT. Now If i get ABAQUS 16 (just a number as an upgrade) which has built-in VUMAT to simulate crack growth with plasticity; i will be very happy.

Another issue is that you cannot do hill's plasticity with GTN damage in ABAQUS; i dont think it can be done readily in any software. So either I write a code, or you write it and I buy or you make a software and solve my two problems and may be I will learn your package because I shall of course be reluctant in learning a new one.

But if you give me readily avaialable Weibull stress, with crack growth simulation and Hills plasticty and many more models built-in, my tendency to get your software will be more. Having said that I would also be concern about your error messaging system and accuracy and of course speed.

Nowadays finite element software includes various components, front-end user interface, solvers, meshers, etc. Making a fully integrated FEA software package could be a huge project.