Model-Based Fault Management

Introduction

Model-based fault detection, fault diagnosis and fault management are not yet widely employed in industry. In many cases, this can be attributed to the fact that the decision makers are not yet aware of the enormous power and benefit stemming from the implementation of model-based fault detection, diagnosis and fault management systems. The following pages shall give a short overview of my research work and of the basic notions.

State of the Art

Currently, only directly available signals are monitored. They are subjected to limit checking, trend checking and/or also plausibility checks of the input signals. These methods are simple, easy to understand and implement. Furthermore, they are very reliable. A scheme for the signal based fault detection and diagnosis is shown in the following figure.

Why is Fault Management Interesting

This conventional approach has many disadvantages, including but not limited to the following:

Thus, a good fault detection and diagnosis system should meet the following requirements:

These requirements are typically met or even exceeded by modern model-based fault detection and diagnosis systems. This brings many advantages:

First, the detection of faults improves the reliability and safety of technical processes, it also increases the availability of plants. Quality control can easily be implemented as the condition of the manufacturing machines is known precisely. Maintenance ressources can be saved as one can constantly monitor the condition of the supervised machines and only initiate counter measures whenever a fault is reported by the machine. This leads to maintenance on demand. The next logical step after the fault detection and diagnosis is the reaction to the fault. This fault management can for example change the configuration of the system (reconfiguration) or switch to a state with less wear and tear ("Limp home" functionality). In this context, the following definitions are often helpful.

Terminology in the Area of Fault Detection and Diagnosis
Fault detection Determination of faults present in a system and time of detection
Fault isolation Determination of type, location and time of detection of a fault, follows fault detection
Fault identification Determination of size and time-variant behavior of a fault. Follows fault isolation
Fault diagnosing Determination of type, location, size and time of detection of a fault. Follows fault detection. Combines fault isolation and fault identification
(Condition) monitoring A continuous real-time task of determining the conditions of a physical system by recording information recognizing and indicating anomalies of the behavior
Supervision Monitoring a physical system and taking appropriate actions to maintain the operation in case of a fault
Protection Means by which a potentially dangerous behavior of the system is suppressed if possible, or means by which the consequences of a dangerous behavior are avoided

General Approach

The following figure shows the general approach to model-based fault detction and diagnosis.

In parallel to the process, there is a process-model which is driven by the processes inputs and outputs as well as additional measurements taken at the process. So-termed features are extracted from the process. These can for example be physical parameters, such as a resistance, an inductivity and such, but also system states or the difference between a model output and a process output. By comparison with the normal process behavior, one can detect changes in the process and thus detect faults. The features which deviate from the normal values are termed symptoms and can be subject to a Symptom-Fault Classification. This allows not only detection but also diagnosis of the fault. Finally, a fault management system can initiate recovery actions. In the case of a sensor fault e.g. the fault management system can switch over to a so-termed model-sensor which supplies an estimate of the actuial measurement for the measured quantity.

Examples

On the following pages, the different notions are explained in a little more depth. Demonstrational videos are included to show how the approaches work in the "real world". The pages are