The methods and algorithms of the PRO-AID code provide for automated diagnosis of equipment faults. Data from plant sensors are sampled periodically and trends are compared against the steady-state condition to determine if an anomaly exists. If an anomaly is detected, the code attempts to identify the cause through a reasoning process that relates faults to sensor trends and knowledge of how plant components are connected.
The code was used to investigate for the Chemical and Volume Control System (CVCS) of a pressurized water reactor (PWR) the sensitivity of fault diagnosis to important parameters. The CVCS provides services to the reactor coolant system (RCS) including maintenance of programmed water level in the pressurizer and maintenance of seal-water injection flow to the reactor coolant pumps. The study investigated the effect on fault diagnosis of parameters such as the type of equipment fault, the severity of the fault, and the spatial coverage of the sensor set. The sensitivity is measured in terms of the fault diagnosis time (time elapsed between fault initiation and diagnosis) and the spatial precision of the diagnosis.
The dynamic response of the CVCS to faults was simulated using the 1-D GPASS systems code for the simplified representation of the CVCS shown in the figure. Equipment faults were introduced by appropriately modifying the physical characteristics of the component. For example, a failure of the CVCS pressure-letdown valve was simulated by changing the pressure loss characteristics of the valve by modifying the loss coefficient for the valve.
Twenty fault cases from the Final Safety Analysis Reports (FSAR) were selected as providing good coverage of transient events that exercise the diagnosis modules in PRO-AID. The twenty cases appear in the accompanying table and represent a variety of components and upset events in the CVCS.
The twenty faults were presented to the PRO-AID algorithm code base with the outcomes then categorized and analyzed for diagnostic assessment. It is not necessarily the case that an exact diagnosis for every fault case can be generated. The configuration of sensors is one of the limiting factors and led us here to categorize diagnostic outcome as complete, semi-complete, or incomplete.
A complete diagnosis means that the outcome of the fault case is correct and localized to the specific fault. In the table, CV20 involves failure of Seal #1; with appropriate sensory information around the seal being passed to the fault diagnosis algorithms, a complete diagnosis is made as illustrated in the figure.
The pressure across the seal and the temperature in and out of the seal are measured which combined with other measurements in the loop yields a seal fault diagnosis (and only the seal). This is the best outcome as far as presenting information to the operator, as it is concise and will aid in the operator’s decisions.
A semi-complete diagnosis means that the outcome of the fault case contains the correct information but may also contain other information not fully localized to the fault (for example, if there are not enough sensors and a region is highlighted, as shown at the bottom of the figure) This outcome is helpful to the operator but may be potentially distracting.
The figure illustrates an incomplete diagnosis that is centered around the CV08 fault in the table where a flow control valve fails. This fault arises from information from pressure sensors, of which there is one at the inlet of CV-121 and one each at the inlet and outlet of the RCP filter (the next component in the loop), as shown in the top of the figure. In this case the diagnosis is complete as the fault diagnosis algorithms have enough information to determine that CV-121 has undergone a malfunction, and the diagnosis reported only centers on this component
However, if the PI-140A pressure sensor and the resulting DP-RCPF1 delta pressure sensor are removed, only leaving the pressure out sensor PI-140B for the RCP filter as shown in the bottom of the figure, then the sensory information presented to the fault diagnosis algorithms does not allow for a complete diagnosis to be made. In this case, the diagnosis is correct in its reasoning and accurate in its placement (given the information presented to the fault diagnosis algorithms), but a larger region of the loop is highlighted.
The results are reported in the table at right. In all cases of incomplete diagnosis, the diagnosis can be changed to complete (or complete to incomplete) by adding (or removing) sensors to the system. With more sensors, the fault diagnosis algorithms had increased visibility within the system and can perform diagnosis based upon this richer sensor set. With fewer sensors, the fault diagnosis algorithms had decreased visibility in the system and the diagnosis will descend towards incomplete outcomes. Therefore, the choice and location of sensors throughout a system and their relation to equipment fault diagnosis must be well understood to ensure a successful deployment of the fault diagnosis algorithms
The characterization of what constitutes an acceptable sensor set is being investigated. See for example Sensor Assignment.