Domain

Explanation

Fault

• Based on Frank's 1994 paper,
• A fault is understood as any kind of malfunction in an actual dynamic system that leads to an unacceptable anomaly (abnormal operating condition) in the overall system performance
• Hence, a fault can be a failure of any form, in any component & within any system
• Two types of faults (Collins et al 2000):

1. Hard failure: abrupt fault with completely cease functioning
2. Soft failure: small or incipient fault whist still working, but not functioning properly - eg. <100% of true output, corruption, fluctuations

Fault diagnosis

• Just as doctors diagnose patients for signs & symptoms of ailments
• Fault diagnosis seeks to find faults (detection) & identify the sources of these faults (isolation)
• Hence, fault diagnosis is also called fault detection & isolation (FDI)
• FDI encompasses the two domains:

1. Theory: algorithms for

• Signal processing techniques
• Logical switching functions

1. Applications:

• Plan of maintenance: repair & replace before faults occur
• Monitoring: of system (plant) to detect faults ASAP & isolate them for control & maintenance
• The plant has the following:

1. Structure: the framework of all in-house processes
2. Sensors: for detection
3. Actuators: detection & compensation

• Compensation can require system reconfiguration in order to continue operations with the new plant
• FDI started with the sensor components, but has now expanded into all plant components

Fault tolerance

• As any system can reasonably expect faults to occur, a good system should be able to withstand certain faults to certain degree
• Hence, the system should be fault-tolerant
• Take for example, a vehicle would be always be equipped with a spare tyre in case of puncture (that is, if only one gets blown)
• That spare tyre is redundant, but acts as a form of fault-tolerant measure against the fault of a puncture
• In R&D, making use of redundancy, fault tolerance consists of:

1. Hardware redundancy: traditional approach; extra similar equipment or system components in excess of the minimum required for normal operations; takes over in fault; disadvantages of higher costs, software, equipment space
2. Functional redundancy: modern model-based approach; extra dissimilar equipment or components; uses sophisticated comparisons & FDI schemes; robustness required

• Hardware redundancy involves:
• Configuration in case of faults to use spare parts & drop faulty ones
• The outputs of the similar parts need to be compared, filtered & by majority vote of similar sensors detect faults
• Enable separate FDI for plant components (slower dynamics) & sensor components (parametric, faster dynamics)
• Functional redundancy involves:
• Dissimilar sensors & components: that work differently but share the same function within the overall system
• Since only function is shared, majority vote techniques not work, hence sophisticated comparison required - FDI schemes

FDI

• 2 types of FDI schemes:

1. FDI based on known or assumed plant dynamics
2. FDI based on online identification schemes

• The objectives are to robustly generate signals for:

1. Fault detection: eg. using instrument fault detection (IFD)
2. Identify faulty component: difficult under feedback conditions as other channels are affected by single faulty component

• To meet these objectives, we need to criteria to gauge the effectiveness of FDI schemes
• Initialise with a list of target faults (fault types) for FDI
• The 5 criteria required for successful FDI are as follows:

1. Promptness: speed
2. Sensitivity: to target faults & insensitive (robust) to all others (eg. other non-target faults, disturbances, noises & uncertainties)
3. False alarm: reduced triggering under non-target faults
4. Missed detection: reduced detection of target faults - use of thresholds & decisions (for diagnosis, observation & redundancy)
5. Incorrect fault identification: wrong identification of target faulty component, resulting in faulty compensation

• It has been consistently emphasized of the critical necessity of robustness against uncertainties:

1. Unmodelled non-linearities
2. Uncertain dynamics
3. Disturbances & noises

• Approaches for robustness:

1. UIOS: eigenstructure assignment using IFD contained in uncertain disturbance distribution à disturbance decoupling à linearity required
2. Model-based using knowledge base of fault types:

• Generating hypothesis: reasoning
• Testing: evaluating hypothesis for truth à approaches like robust filter & SPRT

Disciplines involved

• System identification techniques
• Hypothesis: statistics
• Robust estimation & control
• Optimisation
• Decision logic

Links

• Fault detection control
• Fault isolation control
• Optimal fault detection control
• Optimal fault isolation control
• Optimal fault tolerant control
• Optimal fault compensation control
• Optimal fault control