in Process Control System Presented by:

Transcription

1 Leakage Diagnosis in Process Control System Presented by: Haris M. Khalid

2 Outline Problem Statement Leakage Diagnosis : A critical Issue A proposed Diagnostic Scheme Approaches Employed for Leakage Detection Evaluation of the Proposed Schemes Outcomes from Extensive Experimentation Conclusion

3 Problem Statement Management of leakage faults in fluid system is becoming increasingly important in recent years from the point of view of economy, potential ti hazard, pollution, and conservation of scarce resources.

4 The Bhopal Disaster

5 The Bhopal Disaster The Bhopal disaster was an industrial disaster that occurred in the city of Bhopal, India on December 03, A Union Carbide subsidiary pesticide plant released 42 tonnes of methyl isocyanate (MIC) gas, exposing at least 520,000 people to toxic gases. The Bhopal disaster is frequently cited as the world's worst industrial disaster.

6 the Bhopal Disaster : Cause Large amounts of water entered tank E-610, containing 42 tonnes of methyl isocyanate. The resulting reaction generated a major increase in the temperature of liquid inside the tank to over 400 F (200 C). The MIC holding tank then gave off a large volume of toxic gas, forcing the emergency release of pressure. The reaction was sped up by the presence of iron from corroding nonstainless steel pipelines.

7 Leakage Diagnosis: A Critical Issue

8

9

10 Fault Diagnosis Schemes Model, neural network, and statistical inference based approaches are proposed for the leakage diagnosis. Model based approaches are becoming increasingly gypopular p in recent years. A model of a physical system is highly complex, nonlinear and stochastic, and consequently model based approach tends to be computationally burdensome. In critical applications such as those involving hazardous leaks, it is important to ensure a leak is detected reliably and fast.

11 Proposed Approach for Fault Detection Model-Free Approach: Limit Checking Plausibility Check Knowledge Base Model Based Approach Kalman Filter Parametric Identification

12 Model-Free Based Approach VS Model-Based Approach A model-free based approach is simple and fast but not accurate. A model-based approach is accurate but slow.

13 a combination of two entirely different approaches is employed. The tasks of leakage diagnosis are executed in the order of decreasing importance and increasing precision.

14 Execution of Diagnostics Tasks Limit Checks Increasing Precision Increasing Computation Plausibility Signal Based Method Kalman Filter Increasing Details System Identification

15 Cross-Checking of the Assertion from these approaches confirms Reliability Reduced False Alarm Rates helps to plan Preventive Maintenance Scheme

16 Evaluation of the Proposed Scheme The proposed scheme is evaluated on a process control system in a laboratory interfaced with Lab View. Various Controllers are considered namely On/Off, P, PI and PID.

17

18

19 Two Tank Fluid System: Block Diagram

20

21 Two Tank Fluid System: Modeling Model of DC Motor : di L + Ri = u k ωω dt dω J + b ω = T = k T i dt Model of Pump : Q i = k ω ω Q

22 Model of Tank 1 : dh 1 Qi Qb = A1 dt Model of Tank 2 : dh Qb Q l = A2 dt 2

23 Flow Balance Equations dh1 A1 = Qi Cdb 2 g( H1 H2) dt dh A = C 2( gh H ) C l 2 gh dt 2 2 db 1 2 dl 2 PID Controller

24 Linearizing the non-linear model As our goal is to detect incipient fault i.e. an estimation of small deviations in the state and parameters, a linearized model is employed. H = H + h H = H + h Q Q q 0 i = i + i

25 Linearized Model dh dt 1 = bq ah + ah (a) () dh 2 dt = ah ah (b) where b 1 C =, a = db A 2 2 ( 0 0 g H H ) a C 2 = a1+ dl 0 2 2gH 2

26 Evaluation of the Proposed Scheme: On-Off Controller Knowledge based approach execution Kalman Filter Based Approach Execution Parametric Identification Based approach Execution

27 Knowledge Based Approach Execution The slope of the liquid level is given by: dh A = Q Q i dt l

28 Knowledge Based Approach Execution liquid height and flow rate for varios leakages 2 1 flow time 150 height time Input flow rate & Height under various leakage magnitudes

29 Kalman Filter Based Approach Execution The Kalman Filter is given by: ( ) xk ˆ ( + 1) = Axk ˆ ( ) + Buk ( ) + K yk ( ) Cxk ˆ ( ) rk ( ) = yk ( ) Cxk ˆ ( ) where r (k) is the residual. ( A0, B0, C0) are obtained from (A,B,C) assuming no leakage, that is alpha=0.

30 Kalman Filter Based Approach Execution The flow & Height under no leakage: Note that the height is essentially a Ramp

31 Kalman Filter Based Approach Execution 150 The height profile and the residual of Kalman filter 100 height time residual 5 0 residua al time The Height under no leakage & different The Height under no leakage & different Leakage & the Residual

32 Parametric Identification Based Approach Execution A complete model of the process control system comprising the motor, pump, the tank and the valves are obtained using a parameter estimation scheme. One can know the status of all devices forming the system. The parameters characterizing devices such as flow sensor, level sensor, valves, and pipes are estimated. These estimates help to monitor their status and plan a preventive maintenance scheme.

33 A discrete time model of the fluid system relating the height and the input u takes the form: yk ( ) = ayk ( 1) + ayk ( 2) + buk ( ) 1 2 A recursive least-square square estimation of a, a, b 1 2 ˆ ˆ T θ( n+ 1) = θ( n) + K( n+ 1) y( n+ 1) ϕ ( n) ˆ θ( n) Kk ( + 1) = Pk ( + 1) ϕ( k) ϕ T ( kpk ) ( ) ϕ( k) + 1 T P( k + 1) = I K( k + 1) ϕ ( k) P( k) ˆ θ gives a complete e diagnostic picture of the process control o system including the leakage status, level sensor and flow sensor. 1 is:

34 Parametric Identification Based Approach Execution liquid height and flow rate with leakage flo ow time height time The flow rate & the height under leakage. Note that height is exponential.

35 Evaluation of the Proposed Schemes: P Controller Knowledge based approach execution Kalman Filter Based Approach Execution Parametric Identification Based approach Execution

36 Height Comparison after the implementation of P Controller

37

38 Knowledge Based & Kalman Filter Based Approach Execution: Small Leakage Case

39 Knowledge Based & Kalman Filter Based Approach Execution: Medium Leakage Case

40 Knowledge Based & Kalman Filter Based Approach Execution: Large Leakage Case

41 Knowledge Based & Kalman Filter Based Approach Execution: Very Large Leakage Case

42 Knowledge Based & Kalman Filter Based Approach Execution: all Type Leakage Case

43 Knowledge from the Height Profile It can be seen from the height profiles for small, medium, large and very large leakages, the settling time increases with the degree of leakage. The onset of leakage is indicated by a change in the slope in the height profile.

44 Evaluation of the Proposed Scheme: PI Controller Knowledge based approach execution Kalman Filter Based Approach Execution Parametric Identification Based approach Execution

45 Evaluation of the Proposed Scheme with PI Controller: All Type Leakage Case

46 Evaluation of the Proposed Scheme: PID Controller Knowledge based approach execution Kalman Filter Based Approach Execution Parametric Identification Based approach Execution

47 Evaluation of the Proposed Scheme with PID Controller: All Type Leakage Case

48 Conclusion - The proposed incipient leakage fault diagnosis in process control system using three different approaches was evaluated on both simulated and physical system. The results were encouraging. Leakage diagnosis schemes were robust to measurement noise and system nonlinearity. it The Kalman filter and parameter identification schemes based on approximate linearized model were found to be satisfactory. - However very small leakages were difficult to evaluate as the measurement noise variance was relatively large. The knowledge based approach gives a quick indication of a fault from a change in the slope of the liquid height. Kalman filter indicates a fault by a change in the residual magnitude whereas the parameter identification scheme indicates leakage from the magnitude of parameter estimate. Reliability of the presence of leakage is ensured by using all the three approaches.

49 - The leakage fault is more difficult to detect when the controller is more robust. What is good for controller is not necessarily good for fault detection. - Current efforts are under way to address this problem by considering a combination of more powerful model-free techniques, such as wavelets, and an extended Kalman filter scheme to accommodate the effects of the plant nonlinearity.

50 Question Session Questions should be referred to: Haris M. Khalid Mobile# Ground Line#

51 Thank You