Fuzzy model-based predictive temperature control for nitriding furnace.

摘要

In this thesis, a new Takagi-Sugeno fuzzy model-based predictive control scheme is designed for temperature control of a nitriding furnace. Since performances of such controller mainly rely on the accuracy of the model to represent the plant, a novel method to generate a Takagi-Sugeno fuzzy model from input-output data collected on a real nitriding furnace is developed in order to accurately predict the furnace temperature.;In the first part of this thesis, a new method to generate a Takagi-Sugeno fuzzy model is developed. This fuzzy modeling method is presented with a hierarchical structure including input selection, structure identification and parameter tuning steps. A combination of heuristic, forward and backward techniques is employed in order to model the furnace.;The first step is input selection, in which the main goal is removing dependant inputs and inputs with less contribution to the output, also finding model's order and number of time delays, and eventually, reducing the model's complexity. In this work, this step is carried out in three sub-steps which are finding the main inputs/outputs, delay investigation and selection of the order of inputs relevance. In order to find the main inputs/outputs, a heuristic approach is employed by studying the process data. Then for delay investigation, we suggest a forward selection method employing regularity criterion. This method results in more accurate selections but the number of calculations required for this method is relatively high. By keeping small number of input candidates for this sub-step, the results are achieved rapidly as well as accurately. For selection of the order of inputs relevance, we consider a backward selection method with modifications to fuzzy C-means clustering technique, since it is a relatively fast method. Finally, model inputs are chosen by the accuracy-complexity trade-off.;The second step is structure identification, which aims to establish the relationship between the chosen inputs and output. In this step, we determine the number of fuzzy rules by finding the number of fuzzy membership functions for each input. Here, we employ a branch and bound algorithm.;The task of temperature control, in such furnaces, deals with three different challenges: (i) temperature control accuracy during a typical process cycle; (ii) difference of loads from one batch to another which is affecting the time-delay of the same furnace; (iii) variety of furnaces.;The last step is parameter tuning, which deals with optimization of the fuzzy model. We first investigate the trade-off between prediction accuracy and number of training data pairs used for modeling. Afterward, we investigate the number of learning iterations for model generation in order to reach the optimum prediction accuracy.;For the sake of validation purposes, we compare the results from the new developed model with three models generated with different well-known methods. Results show that this model has the best prediction accuracy with the least complex model structure. Also, it indicates that the new method requires far less computation to reach better results. Moreover, the verification of the methodology indicates consistency in modeling outcomes.;In the second part of this thesis, a new fuzzy model-based predictive controller scheme is generated to track a set-point temperature during nitriding process. This control scheme is able to employ different furnace models for the purpose of temperature prediction without major alterations in controller structure. The new proposed controller has three sub-algorithms. The first one is reference tracking, by searching within possible simulated input/output pairs to find the optimum prediction, compared to a set-point temperature. The second algorithm calculates a correction factor for the accumulated errors during time delay between input and output. The last algorithm adapts the optimum prediction to the correction factor and applies the corrected input to real plant.;In this controller scheme, the Takagi-Sugeno fuzzy model is considered as an external component so the main controller structure remains the same in case of replacing the furnace and its model.;For the performance validation, the new controller is compared with a classical PID controller. The results indicate an overall superiority of the proposed controller over the PID one. The proposed controller has better performances in tracking a set-point, even in the presence of noise and disturbance. Moreover, responses rise and settle faster with a lower overshoot, even in the presence of noise and disturbance. In addition, the proposed controller consumes less overall energy than the PID one.