The multidisciplinary specializations taken into consideration during this research will be Mechatronics (Mechatronics and Robotics domain) and Industrial Informatics (Applied Sciences domain). Both of the specializations mentioned above have a highly multidisciplinary character.
The multidisciplinary character may be tackled by two points of view:
- the curricula have to include disciplines from mechanical science, electric and electronic sciences, computer science, a.s.o. (there are not yet any recommendations stipulated in the national regulations);
- the disciplines have to be divided into fundamental, domain and specialty disciplines (the national regulations recommend a certain ratio, but it isn’t based upon any scientific approach)
Finding the optimal ratio between disciplines and the mathematical model which describes the dependencies between the input variables from both points of view will be the main goal of this work.
A first step will involve a comprehensive data gathering process. The data gathering process will involve the study and analyze of the curricula of Mechatronics and Industrial Informatics specializations from universities around the world. If one cannot find these specializations with exactly these names, similar and/or close names will be taken into consideration. The similarity and/or closeness between specializations will be demonstrated by means of analyzing the curricula.
After the data gathering process, the results will organized as:
- one initial set with the percentage of mechanical sciences, electric and electronic sciences and computer science disciplines for each specialization at each university (this is a first approach, a more comprehensive division may be taken into consideration later)
- another initial set with the percentage of fundamental, domain an specialty disciplines for each specialization at each university.
Each one of this initial sets will be divided into two subsets, one called training subset and one called checking subset. The next step will be to determine a mathematical model which accurately describes the behavior of the input/output data.
Due to the highly non-linear character of the data, methods like linear regression and/or interpolation are ineffective in this case.
A model based upon fuzzy logic and neural networks (neuro-fuzzy) will be used for extracting the laws which should control the data.
After splitting the data into training and checking subsets, a neuro fuzzy model will be built using Matlab software package. An initial fuzzy model will be build (the author will chose the types of membership functions and the number of variations intervals).
The initial fuzzy model will be built by a combined process: it will be generated automatically by the software but it has to be tuned by the authors. Using neural networks and a hybrid backpropagation algorithm the fuzzy model will be trained in order to minimize the errors.
The authors has to choose the total number of training epochs and to stop the training of the neural networks when then training errors reach the minimum.
Finally, the optimized neuro-fuzzy model may be used for representing the mathematical dependencies between the studied input/output data.