Statistical Process Monitoring Using Advanced Data-Driven and Deep Learning Approaches

Theory and Practical Applications

Fouzi Harrou, Ying Sun, Amanda S. Hering, Muddu Madakyaru, abdelkader Dairi

• Publisher: Elsevier
• Year: 2020
• Print ISBN: 9780128193655
• eBook ISBN: 9780128193662
• Topic: Technology & Engineering
• Subtopic: Chemical & Biochemical Engineering
• Index: Technology & Engineering

Statistical Process Monitoring Using Advanced Data-Driven and Deep Learning Approaches tackles multivariate challenges in process monitoring by merging the advantages of univariate and traditional multivariate techniques to enhance their performance and widen their practical applicability. The book proceeds with merging the desirable properties of shallow learning approaches – such as a one-class support vector machine and k-nearest neighbours and unsupervised deep learning approaches – to develop more sophisticated and efficient monitoring techniques. Finally, the developed approaches are applied to monitor many processes, such as waste-water treatment plants, detection of obstacles in driving environments for autonomous robots and vehicles, robot swarm, chemical processes (continuous stirred tank reactor, plug flow rector, and distillation columns), ozone pollution, road traffic congestion, and solar photovoltaic systems.- Uses a data-driven based approach to fault detection and attribution- Provides an in-depth understanding of fault detection and attribution in complex and multivariate systems- Familiarises you with the most suitable data-driven based techniques including multivariate statistical techniques and deep learning-based methods- Includes case studies and comparison of different methods

Chapter 1

Introduction

Abstract

With today's competitive automation environment, demands for efficiency, safety, and high productivity are continuously increasing. Thus, process monitoring is vital for maintaining the desired process performance and specifications. Process monitoring aims to detect potential anomalies that can occur in a monitored process and identify their potential sources. This chapter provides an overview of process monitoring methods. To begin, we present the motivation for using process monitoring, followed by an introduction and a reminder of some of the key definitions, fundamental concepts, and terminology that are used throughout this chapter. We also briefly explain the distinction between different types of faults, such as drift, abrupt, and intermittent faults. In the following section, we discuss the different monitoring methods including model-, knowledge-, and data-based techniques. Finally, we describe the most commonly used metrics for the evaluation of the performance of the different fault detection approaches.

Keywords

Fault detection and isolation; Data-based; Model-based; Supervised and unsupervised methods; Process monitoring

1.1 Introduction

1.1.1 Motivation: why process monitoring

Recent decades have witnessed a huge growth in new technologies and advancements in instrumentation, industrial systems, and environmental processes, which are becoming increasingly complex. Diagnostic operation has become an essential element of these processes and systems to ensure their operational reliability and availability. In an environment where productivity and safety are paramount, failing to detect anomalies in a process can lead to harmful effects to a plant's productivity, profitability, and safety. Several serious accidents have happened in the past few decades in various industrial plants across the world, including the Bhopal gas tragedy [1,2], the Piper Alpha explosion [3,4], the accidents at the Mina al-Ahmadi Kuwait refinery [5] and two photovoltaic plants in the US burned in 2009 and 2011 (a 383 KWp PV array in Bakersfield, CA and a 1.208 MWp power plant in Mount Holly, NC, respectively) [6]. The Bhopal accident, also referred to as the Bhopal gas disaster, was a gas leak accident at the Union Carbide pesticide plant in India in 1984 that resulted in over 3000 deaths and over 400,000 others gravely injured in the local area around the plant [1,2]. The explosion of the Piper Alpha oil production platform, which is located in the North Sea and managed by Occidental Petroleum, caused 167 deaths and a financial loss of around $3.4 billion [3,4]. In 2000, an explosion occurred in the Mina Al-Ahmadi oil refinery in Kuwait, killing five people and causing serious damage to the plant. The explosion was caused by a defect in a condensate line in a refinery. Nimmo [7] has estimated that the petrochemical industry in the USA can avoid losing up to $20 billion per year if anomalies in inspected processes could be discovered in time. In safety-critical systems such as nuclear reactors and aircrafts, undetected faults may lead to catastrophic accidents. For example, the pilot of the American Airlines DC10 that crashed at Chicago O'Hare International Airport was notified of a fault only 15 seconds before the accident happened, giving the pilot too little time to react; this crash could easily have been avoided according to [8]. Recently, the Fukushima accident of 2011 in Japan highlighted the importance of developing accurate and efficient monitoring systems for nuclear plants. Essentially, monitoring of industrial processes represents the backbone for ensuring the safe operation of these processes and to ensure that the process is always functioning properly.

1.1.2 Types of faults

Generally speaking, three main subsystems are merged to form a plant or system: sensors, actuators, and the main process itself. These systems' components are permanently exposed to faults caused by many factors, such as aging, manufacturing, and severe operating conditions. A fault or anomaly is a tolerable deviation of a characteristic property of a variable from its acceptable behavior that could lead to a failure in the system if it is not detected early enough so that the necessary correction can be performed [9]. Conventionally, a fault, if it is not detected in time, could progress to produce a failure or malfunction. Note that there is a distinction between failure and malfunction; this distinction is important. A malfunction can be defined as an intermittent deviation of the accomplishment of a process's intended function [10], whereas failure is a persistent suspension of a process's capability to perform a demanded function within indicated operating conditions [10].

In industrial processes, a fault or an abnormal event is defined as the departure of a calculated process variable from its acceptable region of operation. The underlying causes of a fault can be malfunctions or changes in sensor, actuator, or process components:

• • Process faults or structural changes. Structural change usually takes place within the process itself due to a hard failure of the equipment. The information flow between the different variables is affected because of these changes. Failure of a central controller, a broken or leaking pipe, and a stuck valve are a few examples of process faults. These faults are distinguished by slow changes across various variables in the process.
• • Faults in sensors and actuators. Sensors and actuators play a very important role in the functioning of any industrial process since they provide feedback signals that are crucial for the control of the plant. Actuators are essential for transforming control inputs into appropriate actuation signals (e.g., forces and torques needed for system operation). Generally, actuator faults may lead to higher power consumption or even a total loss of control [11]. Faults in pumps and motors are examples of actuator faults. On the other hand, sensor-based errors include positive or negative bias errors, out of range errors, precision degradation error, and drift sensor error. Sensor faults are generally characterized by quick deviations in a few numbers of process variables. Fig. 1.1 shows examples of the most commonly occurring sensor faults: bias, drift, degradation, and sensor freezing.

Figure 1.1 Commonly occurring sensor faults. (A) Bias sensor fault. (B) Drift sensor fault. (C) Degradation sensor fault. (D) Freezing sensor fault.

We can also find in the literature another type of anomaly called gross parameter changes in a model. Indeed, parameter failure occurs when there is a disturbance entering the monitored process from the environment through one or more variables. Some common examples of such malfunctions include a change in the heat transfer coefficient, a change in the temperature coefficient in a heat exchanger, a change in the liquid flow rate, or a change in the concentration of a reactant.

Thus, sensor or process faults can affect the normal functioning of a process plant. In today's highly competitive industrial environment, improved monitoring of processes is an important step towards increasing the efficiency of production facilities.

In practice, there is a tendency to classify anomalies according to their time-variant behavior. Fig. 1.2 illustrates three commonly occurring types of anomalies that can be distinguished by their time-variant form: abrupt, incipient, and intermittent faults. Abrupt anomalies happen regularly in real systems and...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Preface
• Acknowledgments
• Chapter 1: Introduction
• Chapter 2: Linear latent variable regression (LVR)-based process monitoring
• Chapter 3: Fault isolation
• Chapter 4: Nonlinear latent variable regression methods
• Chapter 5: Multiscale latent variable regression-based process monitoring methods
• Chapter 6: Unsupervised deep learning-based process monitoring methods
• Chapter 7: Unsupervised recurrent deep learning scheme for process monitoring
• Chapter 8: Case studies
• Chapter 9: Conclusion and further research directions
• Index