A self-validating control system based approach to plant fault detection and diagnosis

TL;DR: In this paper, an approach is proposed in which fault detection and diagnosis (FDD) tasks are distributed to separate FDD modules associated with each control system located throughout a plant.

About: This article is published in Computers & Chemical Engineering.The article was published on 2001-03-15 and is currently open access. It has received 31 citations till now. The article focuses on the topics: Control reconfiguration & Fault detection and isolation.

Summary

1. INTRODUCTION

• For the sake of both economy and safety, online process monitoring, fault detection and fault diagnosis have received significant attention in recent years.
• Since the hardware modules associated with these control loops are distributed throughout the plant, it seems sensible to distribute associated detection & diagnosis tasks in a similar manner.
• Considerable research and debate are required before practicable implementations evolve.
• Being steady state based, the concept is independent of any time delays in the plant.
• The focus here is on the two most relevant aspects: on distributing diagnostic tasks to control systems and on those non-distributed methods that might be viewed as having some similarities with the approach described here.

2. REPRESENTATIONAL ISSUES FOR SEVACS KNOWLEDGE GENERATION

• This section examines various ways that cause-effect knowledge can be represented to facilitate its generation.
• The first step is to introduce nomenclature relating to block diagram representations of two standard control systems (Section 2.1).
• These block diagrams are then analysed in Section 2.2 to produce equations that can generate cause-effect knowledge.

2.1 Nomenclature

• The various variables used are defined before going any further.
• Parameter Kc is the proportional gain of the controller and parameters Kv, Kp and Kd are respectively the valve, process and process disturbance steady state gains.
• Note that this structure represents only one form of PID control.

2.2 Generation Of SEVACS Inter-Node Relationships

• This sub-section examines how faults and process disturbances can affect individual control systems, the results are then used to construct SDG representations in the next sub-section.
• A similar approach can be taken for the cascade case.
• In both cases, and for both stable and unstable processes, deviations in [dv] or [dp] will have the same effect on [x].
• The directions of the deviations in the various observations can provide additional information with which to infer the ‘direction’ of the various fault hypotheses e.g. “fails-high” or “fails-low”.

2.3 Representing Control Systems By SDGs

• In Figure 6, the circles around nodes D and E indicate that these nodes are measured; hence node F, which is not circled, is unmeasured.
• Figure 7(A) shows an SDG representation of a typical single loop control system, in which C, V, X and M represent the controller output, the valve opening, the controlled variable and the sensor measurement respectively; θr, dv, dp, dm represent deviations in setpoint, valve bias, process disturbance and sensor bias respectively.
• Individual elements should still be treated separately when performing fault diagnosis.
• This super-node can be analysed 14 using control system related cause-effect knowledge that will be discussed in the next section.
• It is worth pointing out that, as has been discussed, for stability the sign product of any of the control loops in the above SDGs must be ‘−’.

3. SEVACS CAUSE-EFFECT KNOWLEDGE

• Results from the previous section can now be applied to generate tables of causeeffect knowledge, which can be downloaded to the SEVACS.
• The contents of the tables differ depending on whether or not the process has a Type Number of zero.
• Equations (8) — (14) were referred to extensively when deriving this knowledge.
• Tables 2 and 3 describe the various effects that individual faults would have on the observations available for single loop and cascade loop control systems respectively.
• These faults would be addressed by using other approaches.

A sensor bias in a single loop control system or in the outer loop of a cascade control

• If the sensor biases, the controller will take action to compensate for this with 15 the net effect that there will be a deviation in the controller output and the sensor measurement will return to its normal value.
• Both the and the outer loop controllers will attempt to compensate with the net effect that the sensor deviation observed (Ds) will have the same direction as the sensor bias.
• The decision table in the Figure 10 summarises this.
• The direction of the exogenous/ancestor fault or disturbance can then be determined by looking at the following: R*: the relation between a sensor measurement and a controller output; Rex: the relation between an exogenous variable and a sensor measurement; D*: the steady state deviation in a controller output.
• The direction of the valve bias can then be determined by looking at the following factors : Rcv: the relation between the controller output and the valve opening; Dc: the steady state deviation in the controller output.

4.1 Control Systems with Uni-directional Interactions

• A simple set of rules can be derived for those control systems with uni-directional interactions that have the fairly general feature shown in Figure 13.
• If S1 pertains to a Type Number 0 controlled process and its control loop deviates (any element in the control loop deviates), then, initially, the fault candidate will be {S1-sensor-bias, E1, E2, valve-bias in the S1 control loop}.
• There are now two possibilities: 17 S2 is affected: because E1 is the common ancestor of S1 and S2 and according to the fault isolation principle, the fault candidate set shrinks to {S1-sensor-bias, E1}; if the direction of the deviation of S2 contradicts that expected from the S1-sensorbias, {E1} is the only fault route.
• If there is no more information about E2, then these two possibilities can not be separated.
• Otherwise, if E2’s descendants deviate, E2 will be the only fault route.

4.2 Control Systems with Bi-directional Interactions

• Here the controlled variables S1 and S2 affect each other ; either can pertain to a single loop (s.l.) control system or to the inner loop (i.l.) or to the outer loop (o.l.) of a cascade control system.
• There must be at least one common ancestor, which is the fault.
• RS1S22 and RS22S1 represent the relations (or interactions) between the two controlled variables S1 and S22. 4.2.3 Type C Interaction: Inner Loops of Both Cascade Control Systems Interact A Type C interaction is shown in Figure 17: one inner loop controlled variable S12 interacts with the other inner loop controlled variable S22.
• First consider the situation in which the controlled processes here are not capacitive.

5. AN ALTERNATIVE FAULT ISOLATION METHOD FOR INTERACTING CONTROL SYSTEMS

• The procedures described in the last section require different knowledge or rules for different processes.
• Now consider the case where additional knowledge is available e.g. in the form of Figure 23: note first that a common disturbance to F, L and T doesn’t exist and hence L-sensor-bias-high is the only fault that can be diagnosed.
• Both the outer and the inner loop controllers of the temperature control system will deviate, as will CA.
• If Figure 23 is known, the common disturbance can then be replaced with K or K0, and the high-CA can be replaced with high-CA0.

7. CONCLUSIONS

• A self-validating control system based approach to plant fault detection and diagnosis has been proposed that enables the distribution of these tasks throughout a plant.
• The approach itself is targeted on control systems that inherently eliminate steady state error; it is modular, steady state based, requires very little process specific information and should therefore be attractive to control system’s implementers who seek economies of scale.
• Blatantly obvious faults like sticking valves are not accommodated, but these can easily be detected and isolated using a simple rule-base, which can also be distributed to the FDD modules.
• The approach would not be able to detect the presence of a sensor bias if it existed at the time the plant was started up.
• The authors suspicions are that the difficulty, once again, would be more to do with the existence and identification of some form of quasi- steady state, than to revising the approach to accommodate these ‘special cases’.

Figures

Figure 17 so that {RS12S22}, known, then {R*S11S11}, {R*S1 determined by the following: