Model Based Diagnostics/Prognostics
Does your condition monitoring system produce too many false alarms?
Many condition monitoring systems simply monitor the values of individual parameters (e.g. pressure, temperature etc), and generate alarms when a significant increase or decrease in the value of a parameter is detected.
In such systems, the only way to make the system more sensitive to the onset of a fault is to decrease the threshold at which the alarm is generated. The risk inherent in this approach is that by decreasing the threshold, it also becomes more likely that any noise which is present will cause significantly more False Alarms to be generated.
The Solution? Model-based Diagnostics/Prognostics
An alternative approach used by Avenca to overcome the False Alarm vs. Fault Detection Sensitivity trade-off problem is to develop model-based diagnostics/prognostics. This requires processing one of more of the signals being measured by the system, based upon a knowledge (or “model”) of how the fault may be expected to behave. The graphs below show some representative results from developing a model-based approach to detecting faults in a military land vehicle component. The first graph shows that when the vehicle is healthy, the output from the model closely matches the target (i.e. measured) values. However, with the onset of a fault, the predicted values now diverge from the measured ones. It is clear that simply monitoring the magnitude of the measured values would not have detected the fault, and it is only by detecting the divergence between the measured and modeled values that the fault can be detected.
The Benefits?- Reduced False Alarms & Enhanced Diagnostic Capabilities
Since it is unlikely that random noise will produce the particular fault characteristics, using a model-based approach it is possible to make the system more sensitive to the onset of faults without incurring a significant increase in the false alarm rate. The approach can also offer enhanced diagnostics, since it may for example be used to discriminate between different types of fault using a single sensor. If you’d like more details of this approach, then please get in touch.