Observerbased Faultdetection in an Electromechanical Servo
(2004) In MSc ThesesDepartment of Automatic Control
- Abstract
- This master thesis work concerns fault detection and isolation (FDI) of an electro-mechanical servo using analytical redundancy. The purpose is to reduce the number of sensors used in today's fault detection schemes and at the same time keep the level of redundancy in the system. The report will start with a comparison between physical and analytical redundancy. Whichever design method that is selected for the detection scheme, there will be a need for a mathematical model of the servo. The report will show two ways of deriving such a model, one more theoretical and one derived from specifications of the desired servo. The model that eventually was used was the latter one, not that it should matter assuming that the parameters of the real... (More)
- This master thesis work concerns fault detection and isolation (FDI) of an electro-mechanical servo using analytical redundancy. The purpose is to reduce the number of sensors used in today's fault detection schemes and at the same time keep the level of redundancy in the system. The report will start with a comparison between physical and analytical redundancy. Whichever design method that is selected for the detection scheme, there will be a need for a mathematical model of the servo. The report will show two ways of deriving such a model, one more theoretical and one derived from specifications of the desired servo. The model that eventually was used was the latter one, not that it should matter assuming that the parameters of the real servo is known. There is a number of different methods that can be used for FDI. A few different methods, both parameter- and state estimation, will be presented in the report. Among these methods are Fault Detection Filters, Parity space and observer schemes. One of these methods is selected to generate residuals for evaluation by some kind of logic. There is a number of ways to do this evaluation. The two main ideas, threshold logic and statistical tests, will shortly be presented. The method that was finally used was one with an observer scheme and threshold logic. The main reason for this was a desire to keep the FDI as simple as possible. There is a possibility that added complexity will compromise the fault-detection. The reason for this is that a complex system can be harder to tune and thus it might miss detections or detect false detections. The main idea with an observer scheme is to design a number of observers (how many is up to the designer) and feed them with the same reference signal as the plant as well as the plant's output(s). The difference between the plant's and the observers' outputs (the residuals), as well as the observers' internal states, can then be used to detect faults. The idea is to try to find a characteristic pattern in one (or more) residual(s) when, and only when, a specific fault occurs. When these patterns are found (if they exist) it is not that hard to design logic that compares the residual with a predefined threshold-value to determine whether the system is working properly or not. After the design has been explained there will be an evaluation a simulation model of the system. The focus will lie on three main areas: (1) correct detection and isolation of all faults, (2) no false detections of faults when the system is running in normal mode and (3) detection-times. The evaluation shows that a system based on observers and threshold-logic can be used for FDI with good results. All faults have been correctly detected and there have not been any false detections when the system has been running in normal mode. The size of the different detection -times is a matter of parameter adjustment. By changing settings such as the threshold-value in the block 'feltid-räknare' lower detection-times, compared to these presented, can be achieved without compromising the correctness of the detections. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8847987
@misc{8847987,
abstract = {{This master thesis work concerns fault detection and isolation (FDI) of an electro-mechanical servo using analytical redundancy. The purpose is to reduce the number of sensors used in today's fault detection schemes and at the same time keep the level of redundancy in the system. The report will start with a comparison between physical and analytical redundancy. Whichever design method that is selected for the detection scheme, there will be a need for a mathematical model of the servo. The report will show two ways of deriving such a model, one more theoretical and one derived from specifications of the desired servo. The model that eventually was used was the latter one, not that it should matter assuming that the parameters of the real servo is known. There is a number of different methods that can be used for FDI. A few different methods, both parameter- and state estimation, will be presented in the report. Among these methods are Fault Detection Filters, Parity space and observer schemes. One of these methods is selected to generate residuals for evaluation by some kind of logic. There is a number of ways to do this evaluation. The two main ideas, threshold logic and statistical tests, will shortly be presented. The method that was finally used was one with an observer scheme and threshold logic. The main reason for this was a desire to keep the FDI as simple as possible. There is a possibility that added complexity will compromise the fault-detection. The reason for this is that a complex system can be harder to tune and thus it might miss detections or detect false detections. The main idea with an observer scheme is to design a number of observers (how many is up to the designer) and feed them with the same reference signal as the plant as well as the plant's output(s). The difference between the plant's and the observers' outputs (the residuals), as well as the observers' internal states, can then be used to detect faults. The idea is to try to find a characteristic pattern in one (or more) residual(s) when, and only when, a specific fault occurs. When these patterns are found (if they exist) it is not that hard to design logic that compares the residual with a predefined threshold-value to determine whether the system is working properly or not. After the design has been explained there will be an evaluation a simulation model of the system. The focus will lie on three main areas: (1) correct detection and isolation of all faults, (2) no false detections of faults when the system is running in normal mode and (3) detection-times. The evaluation shows that a system based on observers and threshold-logic can be used for FDI with good results. All faults have been correctly detected and there have not been any false detections when the system has been running in normal mode. The size of the different detection -times is a matter of parameter adjustment. By changing settings such as the threshold-value in the block 'feltid-räknare' lower detection-times, compared to these presented, can be achieved without compromising the correctness of the detections.}},
author = {{Bark, Ola}},
issn = {{0280-5316}},
language = {{eng}},
note = {{Student Paper}},
series = {{MSc Theses}},
title = {{Observerbased Faultdetection in an Electromechanical Servo}},
year = {{2004}},
}