Fault detection of multivariable nonlinear systems with uncertainty, using robust nonlinear viewer and neural fault estimator.

Dissertation of master course in electrical-control engineering

Abstract

In this paper, the diagnosis, reconstruction and fault prediction of nonlinear systems with uncertainty are investigated. The diagnostic criterion is the residual signal calculated from the difference between the system output and a sliding mode viewer. The idea of ??sliding mode is to compensate the effect of uncertainties on the residual signal. The ability of the sliding viewer to bring the residual signal to the zero sliding level is in conflict with the sensitivity of the diagnostic method. We solve this problem by adjusting the gain of the sliding viewer. This matching algorithm reduces the gains as much as the problem needs to compensate for model uncertainties. The estimation error reaches the zero level with minimal sliding gain. When the error of the viewer leaves the zero sliding level, the occurrence of fault detection and matching of gain is turned off. Therefore, the effect of the defect in the residual signal is propagated without attenuation by the sliding detector. In addition, the neural network of the estimator is activated to reconstruct the dynamics of the defect in the text of the viewer. The weight update rules help to improve the stability of the method and the convergence of the estimation error. By defining an appropriate destruction threshold for the network weights, the fault prediction algorithm is derived from the weight update rule. Checking the stability of Lyapunov fault detector method in both continuous and discrete time domains. Lyapunov stability proof is rarely seen in past troubleshooting methods. Keywords: troubleshooting, nonlinear systems, model uncertainty, neural network, stability proof. Chapter 1- Introduction 1-1 Preface In manufacturing and manufacturing industries, a lot of effort is spent on producing high quality products. The production of a product with the desired quality should subsequently lead to high safety and attention to environmental regulations. Operations that were once acceptable to us no longer seem suitable due to the rise of our expectations from industries. Therefore, to achieve more favorable standards, in modern industrial processes, several system variables operate under closed loop control. Standard controllers (such as PIDs, predictive controllers, etc.) are designed in such a way as to keep the performance of the system in a satisfactory condition by reducing the effects of disturbance to the system. Although these controllers can handle different types of disturbances, there are some changes that the controller cannot handle. These changes are called defects [[i]]. In other words, any unauthorized deviation in at least one characteristic behavior or parameter of the system can be defined as a fault[1].

The continuous increase in complexity, reliability and efficiency in modern systems requires the continuous development of the field of control and error detection. This requirement clearly shows itself in industries that are critical in terms of safety. These include nuclear power plants, chemical industries and airplanes to new industries such as self-driving vehicles and high-speed trains. Timely diagnosis and identification of errors can prevent the sudden stop of the system and human and financial losses. In figure 1-1. Modern control system describes how to deal with faults in modern systems. As can be seen, the controlled system is the main part of this picture, which includes actuator, sensor and process dynamics. Each of these parts can be affected by external factors such as process noise, measurement noise or external disturbance. In addition, in cases where fault detection with high reliability is discussed, the uncertainties of the system should be considered. In such a situation, the system may still be affected by the defect (as defined earlier [[ii]]. In this case, we expect the fault detection system to be able to separate the fault occurrence from other external factors.

1-1. Modern control system [2]

As previously stated, in general, a fault can be defined as any unauthorized deviation in the behavior or characteristic parameters of the system; For example, the improper performance of the sensor [2] in the system can be considered as a fault. In other words, any unexpected change that degrades the system's performance is included in the field of system defects. In contrast to the defect, the term destruction[3] is also proposed, which refers to the complete stop and collapse of the system.It is worth noting that the defect is more often referred to as improper performance and the use of the term destruction is more appropriate for the occurrence of a disaster; Because, in fact, destruction involves the permanent inability of the device to perform its tasks under defined performance conditions[2]. The classification can be based on the place of occurrence of the fault in the system or based on the time changes of the progress of the fault in the system. Based on the location of the defect, three categories of defects can be defined as follows[2]:

A. Exciter failure[4], which involves malfunctioning of the equipment that excites the system. For example, the fault of the electromechanical drive in a diesel engine.

B. The process defect[5] occurs when changes in the system lead to the invalidity of the dynamic relationships governing the system. For example, tank leakage in a two-tank control system.

c. Sensor fault[6], which manifests itself as serious changes in system measurements.

Also, based on the trend of time changes of the fault, the following categories can be presented[[iii]]:

A. Sudden fault[7], which is modeled as step-shaped functions. This defect usually shows itself as a bias in the evaluated signal.

B. Smooth defect [8], which models it as first-order functions. This defect usually manifests itself in the form of divergence and deviation of the evaluated signal from normal values.

C. Intermittent fault[9] is a combination of shocks with different amplitudes.

In Figure 1-1. The modern block control system [2] under the title of error detection [10] is parallel to the main system. The main role of this block is to monitor the behavior of the system and collect any information related to abnormal performance in any of the system components. Therefore, the task of error detection can be divided into three main parts [2]:

A. Fault detection [11], this part refers to the decision about the state of the system. Detecting that something unusual has happened to the system or that the system is working under normal conditions.

B. Cleaning the fault[12], this section deals with determining the location and location of the fault. For example, which sensor or actuator is involved in the fault.

C. Identifying the defect [13], determining the size, type and nature of the defect has a place in this section.

Different methods of detecting the defect have been designed so far. Also, these methods can be classified into different groups based on different criteria. In this section, the following classification of [[iv]] is provided. Troubleshooting methods can be placed in three different categories:

A. Reliability hardware [14], this method is one of the old troubleshooting methods. The basis of this method is based on the use of several hardware or software sensors, actuators and processors that are responsible for controlling and measuring a particular parameter of the system. Next, a voting system is used to decide on the occurrence and non-occurrence of the fault and the relative location of the fault. It is very common to use this method in very sensitive systems such as flight control. Although this method is very reliable; But additional equipment and their maintenance and repair are costly. In addition, the need for space for the hardware equipment of this method is one of its serious problems.

B. Signal-based methods [15], this method is actually one of the common methods for troubleshooting. The main idea of ??this method is to monitor the level of a specific signal from the system; If this signal reaches a certain threshold, the fault event alarm is activated. This method is very convenient for practical use; But it has its own serious problems and disadvantages. The first problem is that this method is not robust [16]. Not being robust means that in the presence of noise, input changes, or operating point changes, fault event alarms may be activated by mistake. The second problem is that a fault alone can cause a large number of system signals to exceed their thresholds; In this way, it becomes very difficult to detect the location and location of the error. In order to solve these problems, the combination of these methods with statistical and random methods is proposed; This method is to develop the resistance and accuracy of troubleshooting methods.

C. Model-based methods [17], the generality of this method can be expressed as follows: first, we define a mathematical model of the system with the basic information we have about the system; Then some parameters accessible from the main system are measured.