Dynamic Alert Generation Technology for HUM Systems

There are limitations in current alert setting methodologies based on setting fixed thresholds for the amplitude of vibration and other Health & Usage Monitoring – HUM - indicators. HUMS vibration data is both non-stationary and non-ergodic making the performance of classical thresholds, particularly of they are fleet alerts, worse than theory would predict. Alerts set by using a priori statistics exhibit a false alert rate that requires significant man power to manage in order that the false alerts do not translate to an unacceptable false alarm rate and unnecessary maintenance. These classical alerts are also relatively insensitive to defects and result in missed alarms. In general the technology is neither reliable nor robust enough to be used to substantiate maintenance credits. The paper describes two techniques that are extensively utilized in radar and sonar data extraction systems which are beset by similar problems in setting alert thresholds. The techniques are not based on a-priori statistics or assumptions that the data is either stationary or ergodic. The two techniques are: a) The Constant False Alarm Rate (CFAR) technique for setting thresholds based on signal amplitude b) The Box-Car technique for the automatic detection of local trends. These techniques produce dynamic alerts that are based on a significant change in local, not global, conditions in the data stream. The implications of using dynamic alerts, which are relative changes as opposed to fixed absolute levels, are discussed in the paper. The performance of these techniques on real world HUMS generated vibration data and the engineering of the processes to automatically produce reliable and robust alerts are described. The improvements in the performance produced by using these technologies have been demonstrated to be false alert rates of better that 1 in 100,000 and where True Positives can outnumber False Positives by better than 5 to 1. The consequential impact of this level of performance on HUMS dynamic alert processing is assessed. The automation of alert processing will break the link between the number of aircraft fitted with HUMS and the manpower required to process the data. More importantly it will permit the scarce resource of HUMS specialist trained to correctly diagnose defects to be concentrated on that task, not dissipated processing false alerts.