发明名称 System and method for detecting an at-fault combustor
摘要 A system for detecting an at-fault combustor includes a sensor that is configured to sense combustion dynamics pressure data from the combustor and a computing device that is in electronic communication with the sensor and configured to receive the combustion dynamics pressure data from the sensor. The computing device is programmed to convert the combustion dynamics pressure data into a frequency spectrum, segment the frequency spectrum into a plurality of frequency intervals, extract a feature from the frequency spectrum, generate feature values for the feature within a corresponding frequency interval over a period of time, and to store the feature values to generate a historical database. The computing device is further programmed to execute a machine learning algorithm using the historical database of the feature values to train the computing device to recognize feature behavior that is indicative of an at-fault combustor.
申请公布号 US9500563(B2) 申请公布日期 2016.11.22
申请号 US201314097540 申请日期 2013.12.05
申请人 General Electric Company 发明人 Patrick Romano;Lemmon Matthew Francis;Nanda Subrat;White Jonathan David;Pandey Achalesh Kumar
分类号 F23M11/00;G01M15/14;F23R3/00;F23N5/16;F23N5/24 主分类号 F23M11/00
代理机构 Dority & Manning, PA 代理人 Dority & Manning, PA
主权项 1. A system for detecting an at-fault combustor, comprising: a combustor; a sensor configured to sense combustion dynamics pressure data from the combustor; and a computing device in communication with the sensor and configured to receive the combustion dynamics pressure data from the sensor, the computing device programed to: convert the combustion dynamics pressure data into a frequency spectrum; segment the frequency spectrum into a plurality of frequency intervals; extract a feature from the frequency spectrum; generate feature values for the feature within a corresponding frequency interval over a period of time; electronically store the feature values to provide a historical database of the feature values; execute a machine learning algorithm using the historical database of the feature values to train the computing device to recognize feature behavior indicative of an at-fault combustor; convert the combustion dynamics pressure data of a second combustor into a second frequency spectrum; segment the second frequency spectrum into a plurality of frequency intervals; extract a feature from the second frequency spectrum that corresponds to the feature extracted from the first combustor; generate a feature value for the feature from the second frequency spectrum over a period of time, wherein the feature value is generated within the same frequency interval as the frequency interval of the first combustor; and compare behavior of the feature of the second combustor to the behavior of the feature of the combustor; wherein the computing device adjusts operation or indicates that maintenance is required for at least one of the first combustor and the second combustor based on a detection of anomalous behavior between the behavior of the feature of the second combustor and the behavior of the feature of the first combustor is detected, wherein anomalous behavior between the behavior of the feature of the second combustor and the behavior of the feature of the first combustor indicates an at-fault combustor.
地址 Schenectady NY US