Fault Detection & Diagnosis in Control Valve Shahriar iar - - PowerPoint PPT Presentation

Shahriar iar Shahra ram Super ervi visor: sor: Dr. No Noba bakhti hti

BSc Project Fault Detection & Diagnosis in Control Valve

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

2

• f

29

Sharif University of Technology

Content

 What is fault?  Why we detect fault in a control loop?  What is “Stiction”?  Comparing “stiction” with other faults  Ways we can resolve “stiction”  First method : Shape based (MV-OP diagram analysis)  Second method : Cross Correlation Function  Third method : Curve Fitting  Comparing methods  New method : EMD  Challenges

Shahriar Shahram Fault Detection & Diagnosis in Control valve

3

• f

29

Sharif University of Technology

Controller : poor tuning is not fault Actuator Fault: valve friction Plant Fault: leakage , human error Sensor Fault: calibration

What is fault?

Actuator Controller Plant Sensor

SP Output

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

4

• f

29

Sharif University of Technology

Why we detect a fault?

• A. Increasing product quality
• B. Reducing the rate of rejection
• C. Fault signals propagation in physical components
• D. Minimizing risk of instability

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

5

• f

29

Sharif University of Technology

Valve Stiction

What is “Stiction” referred to?

• According to Horch (2000)

‘‘The control valve is stuck in a certain position due to high static friction. The (integrating) controller then increases the set point to the valve until the static friction can be overcome. Then the valve breaks off and moves to a new position (slip phase) where it sticks again. The new position is usually on the other side of the desired set point such that the process starts in the opposite direction again’’.

Reference3

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

6

• f

29

Sharif University of Technology

Stiction in a control valve

Reference 1 Reference 3

 Valve stiction is very important. The first reason is that valve stiction causes 30% of loops work poorly(2nd rank). Another reason is when a valve has stiction , the risk of instability will rise.

Schematic of a control valve

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

7

• f

29

Sharif University of Technology

Comparing “stiction” with Other Faults

Reference1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

8

• f

29

Sharif University of Technology

MV-OP analysis: Method A,B

• This method utilizes the fact that MV does not change even though OP

changes if stiction occurs in control valves.

• We can quantify the degree of stiction by checking the length where

MV stays constant.

No stiction Uncertainty Stiction SIA ≤ 0.25

• SIA > 0.25

Stiction index (SIA)

Reference 1 Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

9

• f

29

Sharif University of Technology

MV-OP analysis: Method C

• Method C is based on the qualitative shape analysis of the characteristics in

Fig.1

• Time segments of signals can be qualitatively approximated by

means of three qualitative symbols: increasing (I), decreasing (D) and steady (S).

No stiction Unceratinty Stiction SIC ≤ 0.25

• SIC > 0.25

Fig.1 Reference 1

Stiction index (SIc)

Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

10

• f

29

Sharif University of Technology

MV-OP analysis: Method A

Advantages

• They are intuitive
• Easy to understand
• Easy to implement
• Computationally efficient
• They work even when no periodical oscillation occurs

Disadvantages

• We should have position of the valve in every moment (only in smart valves)
• Method C is more confident than the other methods but this method also needs choosing an

efficient sample time; Because lowering the sample time increases the noise & increasing the sample time is harmful. [Reference 1]

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

11

• f

29

Sharif University of Technology

CCF : Cross Correlation Function

• If the cross-correlation function (CCF) between controller output and process output is an odd

function (i.e. asymmetric with respect to (w.r.t.) the vertical axis), the likely cause of the

• scillation is stiction. If the CCF is even (i.e. symmetric w.r.t. the vertical axis), then stiction is not

likely to having caused the oscillation.

a)Stiction case b) Non-stiction case

Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

12

• f

29

Sharif University of Technology

CCF : Cross Correlation Function

• Note that the proposed method will work under the following assumptions, which

will be discussed later:

• The process itself is not integrating (such as level control).
• Note that it is important that a procedure for automatic distinction

between odd and even functions needs to have a deadzone.

a)Stiction case b) Non-stiction case

Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

13

• f

29

Sharif University of Technology

CCF : Cross Correlation Function

Non stiction Uncertainty Stiction 0≤∆ρ≤0.072 0.072≤∆ρ≤1/3 1/3≤∆ρ≤1 0≤Δτ≤1/3 1/3≤Δτ≤2/3 2/3≤Δτ≤1

Reference 1

Stiction index (SI)

Reference 1

𝜐𝑠 = 𝑨𝑓𝑠𝑝 − 𝑑𝑠𝑝𝑡𝑡𝑗𝑜𝑕 𝑔𝑝𝑠 𝑞𝑝𝑡𝑗𝑢𝑗𝑤𝑓 𝑚𝑏𝑕𝑡 𝑠0 = 𝐷𝐷𝐺 𝑏𝑢 𝑚𝑏𝑕 0 −𝜐𝑚= 𝑨𝑓𝑠𝑝 − 𝑑𝑠𝑝𝑡𝑡𝑗𝑜𝑕 𝑔𝑝𝑠 𝑜𝑓𝑕𝑏𝑢𝑗𝑤𝑓 𝑚𝑏𝑕𝑡 𝑠𝑝𝑞𝑢 = 𝑡𝑗𝑕𝑜(𝑠0). Max(𝑠

𝑣𝑧(𝜐)) (𝜐∈ [−𝜐1 , 𝜐𝑠 ])

∆ρ = ∣ 𝜐𝑚−𝜐𝑠∣ ∣ 𝜐𝑚+𝜐𝑠 ∣ ∆𝜐 = ∣ 𝜐𝑠−𝜐𝑝𝑞𝑢∣ ∣ 𝜐𝑠+𝜐𝑝𝑞𝑢 ∣

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

14

• f

29

Sharif University of Technology

CCF : Cross Correlation Function

Advantages

• Easy implementing
• Using routine data
• No need to filtering the noise

Disadvantages

• Not practical on integrator systems
• Function phase shift depends on the controller design
• Cross correlation function doesn’t work for dominant proportional controllers [Reference 2]

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

15

• f

29

Sharif University of Technology

Curve Fitting

• To identify stiction-induced oscillations from others, we fit two different functions,

triangular wave and sinusoidal wave, to the output signal of the first integrating component located after the valve.

• OP for self-regulating processes or PV for integrating processes.
• A better fit to a triangular wave indicates valve stiction, while a better fit to a sinusoidal

wave indicates non-stiction.

Not stiction Uncertainty Stiction SI≤0.4 0.4<SI<0.6 SI≥0.6

Stiction index (SI)

Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

16

• f

29

Sharif University of Technology

Curve Fitting

a)Sinusoidal fitting b)triangular fitting

OP/PV signal fitting

Reference 1

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

17

• f

29

Sharif University of Technology

Curve Fitting

Advantages

• One advantage is that it is applicable to both self regulating and integrating

processes.

• Another advantage is its industrial practicability due to the following reasons:

1. The methodology is straightforward and easy to implement.

• 2. The detection is fully automatic and does not require user interaction.

3. Because of the piecewise fit, it is flexible in handling asymmetric or damped

• scillations.

Disadvantages

• Need to know the zero crossings but because of the noise it will be hard.
• Does not guarantee detection of valve stiction in all cases.
• Sufficient data resolution is required for reliable diagnosis.

[Reference 1]

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

18

• f

29

Sharif University of Technology

Comparing

 These methods are the most famous methods because they are :

• Easy to understand
• Easy to implement
• Needs routine data
• Rate of success in analysis
• Curve fitting is the most efficient method among these methods.
• Note that no method works perfectly for all systems.

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

19

• f

29

Sharif University of Technology

New Method : EMD

 EMD or Empirical Mode Decomposition proposed by Norden E. Huang in 1998 for analyzing data from nonstationary and nonlinear processes .  EMD decomposes any time-series signals into the sum of a finite number of Intrinsic Mode Functions (IMFs)  EMD vs. Wavelet Analysis & Fourier Transform  Conditions where the sifting process stops: 1. The residual is a monotonic function 2. The residual has less than two extrema 3. The residual is a constant

Reference 4 Reference 4

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

20

• f

29

Sharif University of Technology

EMD Algorithm

1. Identify all the extrema points in 𝑦(𝑢). 2. Use cubic spline interpolation to connect all the maxima points 3. Compute the average 𝑠

1 𝑢 = (𝑦𝑣 𝑢 + 𝑦𝑚 𝑢 )/2.

4. Compute the fastest oscillation mode (IMF) 𝑑1 𝑢 = 𝑦 𝑢 − 𝑠

1 𝑢

We should continue the steps till the 𝑑1 𝑢 meets the IMF conditions.

• 5. Once 𝑑1 is extracted, the residual 𝑠

1 𝑢

is decomposed as 𝑠

1 𝑢 = 𝑑2 𝑢 + 𝑠 2 𝑢

where 𝑑2 represents the second IMF.

• 6. The separation into IMFs terminate when no further IMF can be extracted.

Reference 4

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

21

• f

29

Sharif University of Technology

Sample of Implementing EMD On a Signal

Original signal

X(t) = 2 sin 2𝜌10𝑢 + 3 sin 2𝜌𝑢 + 0.2𝑢2

IMF1 IMF2 Residual

Reference 4

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

22

• f

29

Sharif University of Technology

Sample of Implementing EMD On a Signal

Reference 5

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

23

• f

29

Sharif University of Technology

Sample of Implementing EMD On a Signal

Reference 5

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

24

• f

29

Sharif University of Technology

Sample of Implementing EMD On a Signal

Reference 5

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

25

• f

29

Sharif University of Technology

EMD: Advantages and Disadvantages

Advantages  Can be used for non-stationary and non-linear data.  Uses only the output signal of the process.  Does not need to define a mother function like wavelet method. Disadvantages × Sensitive to noise × IMF’s are not unique (mode mixing problem)

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

26

• f

29

Sharif University of Technology

EMD: Mode Mixing Problem

X(t) = 2 sin 2𝜌10𝑢 + 3 sin 2𝜌𝑢 + 0.2𝑢2

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

27

• f

29

Sharif University of Technology

EMD: Mode Mixing Problem

X(t) = 2 sin 2𝜌10𝑢 + 3 sin 2𝜌𝑢 + 0.2𝑢2

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

28

• f

29

Sharif University of Technology

Challenges

 Simulating the EMD method

• Advantages and Disadvantages
• How to modify the EMD?

 Extracting IMF specifications for Stiction detection

• Classification
• IMF↔ Stiction

Shahriar Shahram Fault Detection & Diagnosis in Control Valve

29

• f

29

Sharif University of Technology

References

1) Jelali M, Huang B, Detection and Diagnosis of Stiction in Control Loops, Springer New York, 2010. 2) Horch A, Condition Monitoring of Control Loops, PhD thesis, Royal Institute of Technology, Stockholm, Sweden, 2000. 3)

• M. Ale Mohammad, B. Hung, Frequency analysis and experimental validation for

stiction phenomenon in multi-loop processes, University of Alberta, Department of Chemical and Material Engineering, Edmonton, Alberta, Canada ,2011. 4) Ranganathan Srinivasana, Raghunathan Rengaswamya, Randy Miller, A modified empirical mode decomposition (EMD) process for oscillation characterization in control loops, Department of Chemical Engineering, Clarkson University, P.O. 5705, NY 13699, USA bHoneywell Process Solutions, Thousand Oaks, CA, USA, 2007 5) ECG denoising by deterministic approaches, Marjaneh Taghavi Razavizadeh, Sharif University of Technology, International Campus, Kish Island,2014

Questions?