Fuzzy logic based clutch torque curve detection algorithm for heavy duty vehicles

Abstract

In this thesis, a fuzzy logic-based clutch torque curve learning algorithm is proposed as the second method to eliminate the mentioned disadvantages. The torque curve can be determined with this method without the necessity for any specific maneuver and activation conditions. Using a reference point on the curve, the fuzzy logic-based algorithm determines the position value corresponding to the reference point with respect to different clutch temperatures and the first torque transfer points. In this study, 581 Nm was chosen as the reference point. The fuzzy logic theory was introduced by L. A. Zadeh in 1965. Since then, it has been utilized in numerous fields, including the automotive, transportation, robotics, and chemical industries. The theory basically transforms the relationship between concepts into linguistic rules and permits expert opinions and experiences to be incorporated into system models. Fuzzy controllers consist of three main parts: fuzzifier, rule-based inference engine, and defuzzifier. Mamdani and Takagi-Sugeno type of fuzzy controllers are the most commonly used. MATLAB-Simulink was used for simulation studies. First of all, the conventional algorithm model was developed. The activation conditions, timer, and curve calculation functions used in the model are mentioned in detail. Secondly, two different fuzzy controllers, Takagi-Sugeno and Mamdani types, were designed. The purpose of designing different types of controllers is to compare the performances of the controllers for this problem. While designing the controllers, MATLAB's "Fuzzy Logic Designer" interface was utilized. In order to make a realistic comparison, the same input membership functions and rules are used in the controllers. The inputs of the controllers are selected as the clutch temperature and the first torque transfer point. Three membership functions are defined for each input: "low", "medium" and "high". The output of the controllers is the clutch position corresponding to the reference torque. As with the inputs, three different output membership functions are defined as "low," "medium," and "high" for both controllers. During the design of fuzzy controllers, the relationship between inputs and outputs was determined by analyzing data collected from multiple vehicles. After designing both controllers, a mechanism was created to choose between the conventional algorithm and the fuzzy-based algorithm. The decision mechanism basically compares the reference clutch position values obtained from the two strategies. If the difference between the calculated reference values exceeds a predetermined upper threshold, the error is detected, and the curve obtained from the fuzzy-based strategy becomes equal to the final output. If the difference between the calculated reference values is below a lower threshold, the error is deactivated, and the curve obtained from the conventional algorithm becomes equal to the final output. Thus, as the traditional algorithm will not be activated until the first launch maneuver, the error value will be high and the fuzzy-based strategy will be effective. So, the mechanism eliminates the feeling of poor performance on the first launch. Moreover, the output of the fuzzy controller will be continuously updated based on the change in clutch temperature and the first torque transfer point while driving. The fuzzy controller will be activated if an error is detected, preventing incorrect torque curve learning situations. For testing and validating the developed model, a two-step test procedure was created. First, launch maneuver data was collected for three different clutch temperature ranges: low (40-70°C), medium (70-90°C), and high (90-120°C) from a test vehicle with a 28-ton, construction truck variant. The relationship between traditional and fuzzy controller-based algorithms was examined by feeding the vehicle data to the generated MATLAB-Simulink model. This study was carried out separately for models using Takagi-Sugeno and Mamdani type fuzzy controllers. The obtained clutch torque curves were compared for 40, 70, and 100 °C clutch temperatures, one value from each temperature zone. In the second step of the test, the torque curves obtained from the conventional algorithm, Mamdani, and Takagi-Sugeno type fuzzy controllers for different clutch temperatures were validated by performing launch maneuvers on the same test vehicle. For each test, the maneuvers were repeated with the same gear, accelerator pedal, and road conditions. The verification was done by examining the difference between engine and clutch torque during the launch maneuver. A large difference between torque values indicates that the clutch is in the wrong position. For this reason, the difference between the torque values was defined as the error. Three different performance indexes ISE, ITSE and ITAE were used to compare the performance of the strategies analytically. Since the ITSE and ITAE indices are time-dependent, they evaluate launch maneuvers in terms of duration. The test results were analyzed in three sections as low, medium, and high. At low clutch temperatures, both Mamdani and Takagi-Sugeno fuzzy controllers outperform the conventional algorithm. Moreover, Mamdani provides better results according to ISE index, whereas Sugeno outperforms according to ITAE and ITSE indices at low clutch temperatures. The main reason for this is that when a Sugeno-type fuzzy controller is used, the launch times are reduced. For medium clutch temperatures, all three strategies were yielded similar results. As at low temperatures, Mamdani provides better results according to ISE index, whereas Sugeno outperforms according to ITAE index at medium clutch temperatures. According to the ITSE index, the performance of the two strategies is equal. For all three indices, the traditional algorithm has the lowest performance. However, there is no dramatic difference in the results of the three strategies. For high clutch temperatures, Sugeno has the worst performance according to all three indices. The main reason for this is that the Sugeno type fuzzy controller is much more sensitive to high clutch temperatures than the Mamdani type fuzzy controller. In addition, Mamdani type fuzzy controller has the best performance for all three indices. In general, it was observed that fuzzy controllers improved clutch torque curves. On the other hand, fuzzy controllers increased computational load and simulation times. Both types of fuzzy controllers have improved the performance of the first launch maneuvers. Sugeno type fuzzy controller is highly sensitive to changes in high clutch temperatures. Therefore, it showed poor performance at high temperatures. The Mamdani-type fuzzy controller, on the other hand, succeeded in all three test scenarios.