Hi. I'm generating the amount of data required for training an artificial neural network (ANN) using a reliable and validated self-developed numerical code. Is it right approach?
Or the necessary data should be produced only with experimental tests ?
Best Regards
Saeed
you can simulate the model in Matlab and generate data for your work and project. If your project have mathematical model you can use this technique.
The results of the ANN cannot be more accurate than your model, probably less.
Sometimes people do this because the ANN approximation will be quicker than the numerical code - it provides a cheap way to get an approximate answer.
I suggest you to use a simple Spring Damper configuration in ADAMS software to generate displacement velocity and acceleration easily
and train your proposed Neural Networks with these datas.Because Adams is a Multi sensor dynamic Environment ,can help you to simulate like a real experiment and generate huge number of different types of datasets.
Thanks for your good points.
My developed code is on the basis of transient coupled heat and mass transfer equations.
Dvelopment of the code spends many efforts. Thus predicting the phenomonon using the ANN model appears to be precise.
But my main concern was about the way of producing data for handling it. As you said, I can employ numerical results for this purpose.
When I was surveying the literature, most of them used sets of experiments as the training data.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training