Academia.eduAcademia.edu

Figure 1 – uploaded by Angelo Beltran Jr.

See full PDFdownloadDownload figure
decade on various fields such as robotics, industrial control, agricultural, household automation, plant growth modeling, data mining, etc. Notable studies have been made in the past years which use adaptive neuro-fuzzy inference system (ANFIS). Examples are fault diagnosis in power transformers [2], speed control of a switched reluctance motor (SRM) for various real time applications [3], and variable temperature analysis and monitoring [4]. The back-propagation algorithm for calculating a gradient is a special case of a general technique called differentiation in the reverse accumulation mode. Fig. 1. Scheme of the back-propagation algorithm in ANN [5].

Figure 1 decade on various fields such as robotics, industrial control, agricultural, household automation, plant growth modeling, data mining, etc. Notable studies have been made in the past years which use adaptive neuro-fuzzy inference system (ANFIS). Examples are fault diagnosis in power transformers [2], speed control of a switched reluctance motor (SRM) for various real time applications [3], and variable temperature analysis and monitoring [4]. The back-propagation algorithm for calculating a gradient is a special case of a general technique called differentiation in the reverse accumulation mode. Fig. 1. Scheme of the back-propagation algorithm in ANN [5].

Related Figures (8)

Fig. 4. Conceptual framework structure. The approach to integrate neural and fuzzy systems is through cooperative fuzzy neural network implementation. In the case of cooperative neural fuzzy systems both artificial neural network and fuzzy system work independently from each other. The ANN learns the parameters from the fuzzy system.
Fig. 4. Conceptual framework structure. The approach to integrate neural and fuzzy systems is through cooperative fuzzy neural network implementation. In the case of cooperative neural fuzzy systems both artificial neural network and fuzzy system work independently from each other. The ANN learns the parameters from the fuzzy system.
In the detector, there are reference and measuring chambers in each of which a thin platinum wire is stretched. The reference chamber is filled with reference gas and through the measuring chamber, sample gas is flowed. Each platinum wire composes a bridge circuit in combination with an external fixed resistor and it is heated by a flowing constant current. The resulting thermal change is taken out as a change in electric resistance according to which the concentration of measured gas is calculated. When all four resistances are the same Vour is zero, and the bridge is considered balanced. When zeroing, the reference gas is passed over all the filaments, the resistances will be the same (because filaments are matched) and the bridge is balanced. When the sample gas is passed over half of the bridge, then Vour value correlates to the content of the sample gas in the reference. The detector is a four element katharometer having two elements situated in the reference gas, and two elements in the sample gas as shown in Fig. 3.
In the detector, there are reference and measuring chambers in each of which a thin platinum wire is stretched. The reference chamber is filled with reference gas and through the measuring chamber, sample gas is flowed. Each platinum wire composes a bridge circuit in combination with an external fixed resistor and it is heated by a flowing constant current. The resulting thermal change is taken out as a change in electric resistance according to which the concentration of measured gas is calculated. When all four resistances are the same Vour is zero, and the bridge is considered balanced. When zeroing, the reference gas is passed over all the filaments, the resistances will be the same (because filaments are matched) and the bridge is balanced. When the sample gas is passed over half of the bridge, then Vour value correlates to the content of the sample gas in the reference. The detector is a four element katharometer having two elements situated in the reference gas, and two elements in the sample gas as shown in Fig. 3.
Table 2. ANN sample cases.
Table 2. ANN sample cases.
Fig. 5. Sample training and output performance of back-propagation with fuzzy rules. Plots on Fig. 5. shows best performance of the ANN after several attempts with the case samples. Vertical axis represents training error (Mean Square Error, MSE) while the horizontal axis represents the number of learning. Multiple trials and case are established to find the model with the least number of iterations possible. The result is then compared to the ANFIS.
Fig. 5. Sample training and output performance of back-propagation with fuzzy rules. Plots on Fig. 5. shows best performance of the ANN after several attempts with the case samples. Vertical axis represents training error (Mean Square Error, MSE) while the horizontal axis represents the number of learning. Multiple trials and case are established to find the model with the least number of iterations possible. The result is then compared to the ANFIS.
Fig. 6. (a) Membership rules (b) FIS editor (c) Membership function editor
Fig. 6. (a) Membership rules (b) FIS editor (c) Membership function editor
Table 1. Gas measurement training samples. Table 2 shows the sample cases used for the ANN simulations using MATLAB.
Table 1. Gas measurement training samples. Table 2 shows the sample cases used for the ANN simulations using MATLAB.
Fig. 10. (a) ANFIS model structure (b) Surface results of sensor versus flow rate (c) Surface results of sensor versus pressure
Fig. 10. (a) ANFIS model structure (b) Surface results of sensor versus flow rate (c) Surface results of sensor versus pressure
Related topics:
Thermal EngineeringThermodynamicsNumerical SimulationsSimulation (Computer Science)Fuzzy set theoryProcess Modeling and SimulationNeural Control of MovementModeling and SimulationOil and Gas LawGAS TURBINE
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved