Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015 978-4799-8422-0/15/$31.00©2015 IEEE Adaptive PID Controller based on Lyapunov Function Neural Network for Time Delay Temperature Control Muhammad Saleheen Aftab and Muhammad Shafiq College of Engineering Sultan Qaboos University Muscat, Sultanate of Oman saleheen.aftab@gmail.com, mshafiq@squ.edu.om Abstract—Temperature is an important control variable in industrial processes. In this paper, an adaptive PID control algorithm has been discussed to track the process temperature. The presented control algorithm employs Lyapunov function based artificial neural networks for online tuning of proportional, integral and derivative actions. This algorithm has been successfully tested on the laboratory temperature control process trainer. For comparative analysis, the results have been contrasted with the conventional PID scheme. The experimental findings show that improved and stable tracking is achieved with the proposed adaptive PID controller. Keywords—adaptive PID control; Lyapunov function neural network; first order time delay systems; error backpropagation; PID tuning I. INTRODUCTION Temperature is an important control variable in food processing, chemical reactions, power production, heating, ventilation and air-conditioning (HVAC) systems and metallurgical industries. Temperature control is necessary to maintain product quality and equipment safety [1-3]. However, most thermal processes are inherently nonlinear and exhibit uncertainties and time delays [4, 5]. Consequently, an appropriate control design becomes a challenging task. PID controllers are most commonly used in industries due to simple structure and ease of implementation [6]. However, PID controllers tuned with conventional methods exhibit poor performance for time-delay systems [7-9]. New PID tuning algorithms have been proposed [10, 11], yet their performance degrades with variable time delays. Recently, many algorithms have been developed based on fuzzy logic inference mechanism [12-15]. But in most fuzzy systems, membership functions and their intervals must be known in advance, which is not always possible [16]. On the other hand, artificial neural networks are capable of approximating nonlinear dynamic systems [17] and have been successfully employed in adaptive control techniques [18-20]. In this paper, we introduce an adaptive PID algorithm which is based on Lyapunov function neural network tracking (LNT) controller [21]. In this technique, neural network is trained online with a Lyapunov function backpropagation learning algorithm, which guarantees error convergence and closed loop system stability. The proposed algorithm employs the LNT to adaptively tune the proportional, integral and derivative action on the closed loop error. In this work, we have tested the proposed controller on a temperature control process trainer (PT326) and achieved smooth adaptive tracking of desired temperature profile. Experimental results suggest that this technique is superior in performance as compared to the conventional PID control. The rest of this paper is organized as follows: In section II, a brief description of the PT326 process trainer along with its calibration and open loop performance is presented. This section also presents the problem statement. Section III discusses the proposed adaptive PID control technique. In section IV, the experimental results are presented and analyzed. Section V concludes the paper. II. PRELIMINARIES AND PROBLEM STATEMENT A. System Description The Process Trainer PT326, manufactured by Feedback Instruments, exhibits the functionality of many industrial thermal processes. This laboratory apparatus, shown in Fig. 1, consists of an electrically driven fan, which induces air flow through a duct. The incoming air is heated by a voltage controlled heater placed at the beginning of the duct. The heater power is obtained using a linear power amplifier and the output power of this amplifier is proportional to its input voltage. From here onwards, this input voltage is referred as the control input. A throttle is located above the fan which can be used to adjust the air flow manually. The angular movement of the throttle is limited between 10° (minimum) and 170° (maximum). A thermistor measures the temperature of the heated air. It can be placed at three different positions marked Position 1, Position 2 and Position 3 on the duct, as shown in Fig. 1. The sensor positions 2 and 3 introduce time-delay in the system output. The conditioned output voltage of the thermistor is available at terminal Y whereas the analog temperature scale shows the temperature in °C. The trainer comes with a built in proportional controller that can be used in