Intelligent Techniques for Control and Fault Diagnosis in Pressurized

Water Reactor: A Review

SWETHA R. KUMAR, JAYAPRASANTH DEVAKUMAR

Department of Instrumentation and Control Systems Engineering,

PSG College of Technology,

Coimbatore-641004, Tamil Nadu, INDIA

effective, Artificial Intelligence (AI) has grown implausible in recent times. This review article discusses on few

AI techniques to optimally control the neutron flux density and design an effective fault diagnosis algorithm to

detect sensor faults in the nuclear reactor core.

future guidelines enacted to satiate the world's

energy demand. Computational intelligence has long

been considered to have a variety of uses in the

nuclear sector. Artificial intelligence has progressed

tremendously in recent years due to computing

power and cheaper hardware developments.

Approximately 10% of the world's electricity is at

present produced by nuclear reactors, and more

nuclear power stations are enthusiastically being

built across the globe. However, the nuclear sector

algorithms like Neural Networks [2,3], Genetic

Algorithms [4,5,6], Particle Swarm Optimization

[7,8], Ant Colony Optimization [9,10], Artificial Bee

estimation and fault detection in Pressurized

Table 1: Abbreviations

Table 2 Operable Nuclear Power Plants

in Figure 1.5 [18], steam preserves the pressure.

Table 3 Values of parameters used in reactor model

research areas in the domain of Artificial

Intelligence, which has become prevalent in solving

territories or other animal communities [104].

inputs u* in obtaining the optimal output y* with

minimum cost J*. Optimization problem is solved

by choosing the design parameter, then formulating

the constraints and defining a cost function. The aim

of the cost function is to determine a value for

chosen design parameter satisfying the given

constraint that delivers the optimum response. With

respect to controller tuning, optimization algorithms

will provide optimal controller tuning parameters

controller comprises three components namely:

proportional gain kp, integral gain ki and derivative

gain kd. The derivative gain develops the

control action according to the rate of change of

error, the integral gain develops the control action in

response to the sum of all past errors, and the

proportional gain produces the control action for

Swarm Intelligence algorithms mimic the

Figure 2 General Framework of Swarm Intelligence

Algorithm

Algorithms

system [29]. In this work, a novel cost which is a

function of Overshoot, rising time, settling time,

steady state error, Gain Margin (GM) and Phase

drops or changes in loading, the proposed controller

is employed in speed control loops. The simulation

results demonstrate the performance and efficacy of

the suggested strategy for frequency decrease or

implemented and tested on several plants with

promising results that compare with known methods.

system performance. Youssef Dhieb et al. employed

an ACO-tuned PID to eliminate the induction motor's

harmonics and speed ripple [37]. Using the finite

element method (FEM), the parameters of this motor

this method.

display a conventional frequency spectrum.

Every departure from this is a sign of

abnormality.

information is analyzed qualitatively using

tools such as if-then rules. These qualitative

approaches can process inaccurate and

inadequate information to reach a fault

decision. Two prevalent techniques are

expert systems and fuzzy logic.

training data attained during standard

operations. Thus, these methods are also

denoted as process history-based

approaches. Principal Component Analysis,

Artificial neural networks are extensively

utilized for Fault diagnosis [93, 96, 97].

4.2 Model-based Approaches

model-based fault diagnosis methods. In model-

based FDD, the standard behaviour of a process is

states. Residuals are the variations between the

analytically predicted values and the

true measurements. Faults lead to violations of the

normal relationships represented by the model,

which causes the residuals to fluctuate abnormally.

Therefore, faults can be found by statistically