Hardcover: 504 pages
This book provides a detailed treatment of model-based fault detection and diagnosis methodology, using the consistency (parity) relation approach. Much of the material is based on the authorís own research.
The First Chapter gives a general introduction to the subject, with two application examples (automobile engine and chemical plant diagnosis). Chapter Two is a review of the tools of the analysis and design of linear discrete dynamic systems. Chapter Three summarizes some of the fundamental concepts in probability and statistics. Basic systems identification methodology is reviewed in Chapter Four.
The main subject matter starts in Chapter Five, with the introduction of analytical redundancy, the most important concept in model-based fault detection and diagnosis. Emphasis is focused on the generation of residuals, quantities that are nominally zero and that become nonzero in response to faults in the system. The main difficulty of detection is that residuals are also influenced by nuisance effects, such as noise, disturbances and model errors. A further issue is the manipulation of residuals so that they facilitate fault isolation, that is, the determination of the faultís location. Chapter Six discusses in much detail the generation and manipulation of single residuals, from the transfer function and the state-space model of the system, to satisfy certain design specifications. †Chapter Seven is devoted to the design of structured sets of residuals where individual elements are selectively sensitive to subsets of faults, resulting in fault-specific response patterns. In Chapter Eight, directional residual design is addressed, whereas residual vectors lie in specific directions in response to individual faults. While treatment in the previous chapters is limited to faults acting additively in the system equations, Chapter Nine extends the concepts and methods to multiplicative faults, those affecting the parameters of the system.The last four chapters are devoted to special subjects. Chapter Ten addresses robustness in residual generation, in the face of additive disturbances and model errors. Methods to improve robustness are introduced and their limitations outlined. The subject of Chapter Eleven is statistical testing of the residuals; the generalized likelihood ratio approach is applied to the residuals, under the constraints embodied by directional or structured design. Chapter Twelve addresses some problems of model identification for the diagnosis of additive faults while Chapter 13 is devoted to the diagnosis of parametric faults by systems identification methods.
The book closes with a list of 191 references.
Excerpt from the book
Ever since humans have been building machines, they have been naturally concerned about their condition. For centuries, the only way to learn about malfunctions and their location was by biological senses (an approach still practiced); looking for changes in shape or color, listening to soundsunusual in strength or pitch, touching to feel heat or vibration, and smelling for fumes from leaks or overheating. Later, measuring devices were introduced, which provided more exact information about important physical variables. However, these devices (sensors) also proved prone to malfunction, raising the dilemma of false alarms. The potential for faults in the sensors became even more critical when they were applied in the automatic control of the machines, where the effects of such malfunctions may be more direct and devastating, and where the human operator is frequently removed from the process.