Sensor failures include four main categories: complete failure failures, fixed deviation failures, drift deviation failures and accuracy degradation.
Failure failure refers to the sudden failure of the sensor measurement, the measured value has been a constant; deviation failure mainly refers to the sensor's measured value and the true value of a constant difference between a class of faults, as seen in the figure, there is a fault in the measurement of the measurement of the measurement is parallel to the measurement of no fault;; the
Drift faults are faults in which the difference between the sensor's measured value and the true value increases over time.
Accuracy degradation refers to the deterioration of the sensor's measurement capability and low accuracy. When the accuracy level decreases, the average value of the measurement does not change, but the variance of the measurement changes.
Fixed deviation faults and drift faults are faults that are not easy to detect and cause a series of unanticipated problems in the course of the fault, making the control system unable to function properly for a long period of time.
Sensor failure classification way
1, according to the degree of sensor failure classification
According to the size of the degree of sensor failure can be divided into hard failure and soft failure.
Hard failure refers to the structure of the damage caused by the failure, the general amplitude of large, sudden changes; soft failure refers to the characteristics of the variation, the amplitude is small, slow changes.
Hard failure, also known as complete failure, complete failure when the measured value does not change with the actual change, always maintain a certain reading. Usually this constant value is usually zero or the maximum reading. The measured value of the fault is roughly a horizontal straight line.
Soft faults include data deviation, drift, and degradation of accuracy levels. Soft faults are relatively small, difficult to be found, so, in a sense, soft faults harm than hard faults harm is greater, and its harm has gradually attracted attention.
2, according to the failure of the performance classification
According to the performance of faults can be divided into intermittent faults and permanent faults.
Intermittent failure is good or bad; permanent failure failure, can not be restored to normal.
3, according to the failure, the development of the process of classification
According to the process of fault occurrence, development can be divided into mutation fault and slow change fault.
Mutant fault signal rate of change is large; slow-change fault signal rate of change is small.
4, according to the cause of the fault classification
According to the cause of the fault can be divided into deviation faults, impact faults, open-circuit faults, drift faults, short-circuit faults, periodic interference, nonlinear dead zone faults.
The causes of deviation faults are: bias current or bias voltage, etc.; and
Fault causes of inrush faults are: random disturbances in the power supply and ground, surges, spark discharges, burrs in the D/A converter, etc.; and
Fault causes of open-circuit faults: broken signal lines, chip pins are not connected, etc.
The cause of drift faults: temperature, etc.; short-circuit faults: contamination.
Fault causes of short-circuit faults: bridge corrosion caused by pollution, line shorting, etc.
Cyclic interference failure causes: power supply 50 Hz interference, etc.;; and
Fault causes of nonlinear deadband faults: amplifier saturation, containing nonlinear links, etc..
In addition, from the point of view of modeling and simulation, it can be divided into multiplicative and additive faults. For bias faults, the original signal plus a constant or random small signal; for shock interference, can be superimposed on the original signal a pulse signal; for short-circuit faults, the signal is close to zero; open-circuit faults, the signal is close to the sensor output maximum; drift faults, the signal at a certain rate offset from the original signal; cyclic interference faults, the original signal is superimposed on the signal of a certain frequency.
Sensor fault diagnosis methods
From different perspectives, the classification of fault diagnosis methods are not exactly the same. Fault diagnosis methods are simply divided into: methods based on analytical mathematical models and methods that do not rely on mathematical models.
1. Methods based on analytical mathematical models
According to the different forms of residuals, the methods based on analytical mathematical models can be further divided into: parameter estimation method, state estimation method and equivalent space method.
The model-based fault diagnosis method is one of the earliest diagnostic methods developed, but also one of the most widely studied and applied diagnostic methods.
The advantages are that the model mechanism is clear, the structure is simple, easy to realize, easy to analyze, and can be diagnosed in real time. It has an important position in the field of fault diagnosis, and will still be the main research direction of sensor fault diagnosis methods in the future development.
The disadvantages are the large amount of computation, system complexity; the existence of modeling errors, poor adaptability of the model; poor reliability, prone to false alarms, omissions and other phenomena; the robustness of external perturbations, the system is not sensitive to noise and interference.
At present, the research results of this diagnostic method are still mainly focused on linear systems, which is of great significance for the in-depth study of generalized fault diagnostic techniques for nonlinear systems, and at the same time, the problem of robustness is also of high research value. Table l describes the advantages and disadvantages of some fault diagnosis methods in the modeling method.
2. Fault diagnosis methods that do not depend on mathematical models
Currently, the control system becomes more and more complex, due to the fact that it is difficult to establish an accurate analytical mathematical model of the control system in practice, when there is a modeling error, the model-based fault diagnosis methods will be false alarms, omissions and other phenomena, so the model-independent fault diagnosis methods have been highly valued.
The advantages of the mathematical model-independent methods are that they do not require an accurate model of the object and are highly adaptable. The disadvantage is that the structure is complex and difficult to realize.
Such system model-independent fault diagnosis methods can be categorized into fault diagnosis methods based on data-driven approaches, knowledge-based fault diagnosis methods, and discrete-event-based methods.
2.1 Data-driven methods
There are two main categories of data-driven methods: signal processing methods and statistical methods.
Some commonly used signal processing based fault diagnosis methods are: absolute value test and trend test, fault detection using Kullb ack information criterion, fault detection methods based on adaptive sliding Lattice filter, fault detection methods based on signal modal estimation correlation analysis methods, wavelet analysis methods and information fusion methods.
2.2 Knowledge-based methods
Knowledge-based fault diagnosis methods can be concordantly categorized into two types: symptom-based fault diagnosis methods and qualitative model-based fault diagnosis methods.
2.3 Discrete event-based methods
Discrete event-based fault diagnosis method is a new type of fault diagnosis method developed in recent years. The basic idea is that the state of the discrete event model reflects both the normal state and the fault state of the system.
With the progress of theoretical research and the continuous improvement of the technical level, the study of sensor fault diagnosis will tend to be more practical, and some of the problems encountered in practice will be gradually solved.