The paper describes a new learning algorithm of adaptive neuro-fuzzy inference systems that is based on the method of areas’ ratio (MAR-ANFIS). Using linear and nonlinear functions we obtain a generalized model for fuzzy inference. Considering various implication methods, different t- or s- norms and equations for fuzzy inference composition we can change the properties of the resulting output variable. As an example, we illustrate the proposed learning algorithm and show its distinctive characteristics. Firstly, MAR-ANFIS learning algorithm is additive. Secondly, soft operators provide symmetry for the output variable. Also, the proposed algorithm that allows improving accuracy when learning fuzzy system and speed of its learning. Using detailed numerically calculated RMSE and MAPE we evaluate the proposed algorithm. High accuracy of the proposed MAR-ANFIS is confirmed through the calculation of the learning time of neuro-fuzzy network RMSE and MAPE.
Heating of a cutting tool during the processing detail on CNC machine cause thermal deformations which reduce the quality of the machined surfaces. The authors propose to use Peltier thermoelement to control thermal deformations. The article presents two devices to control thermoelement based on a current generator and a field effect transistor. Both devices change the current signal transmitted to the thermoelement by using two fuzzy-logical models. We used FPGA Spartan 3E for increasing speed of fuzzy models and the both devices. Therefore, the main goal of the article is to increase the speed of information processing in fuzzy devices. We increased the speed of processing information by using singleton membership functions on the output of fuzzy models, optimizing and parallelizing the program code for calculating fuzzy operations and replacing the division operations by shifts. Time tests conducted at frequency of 200 MHz showed that the current was calculated to 380 ns, the voltage conversion was carried out to190 ns.