Predictive maintenance oriented pattern recognition system based on ultrasound data analysis

Ultrasonic data analysis as a method of early-stage machine fault detection has been successfully utilized in the Predictive Maintenance (PdM) industry for over 25 years. The challenge with this type of machine health monitoring is the sensitivity of the data collected to the changes in machine operational condition such as load, speed and/or pressure. Typically any alarm or fault criteria threshold based on ultrasonic data would need to be sufficiently elevated in order to accommodate the machine's variable operational conditions or the monitoring system would generate excessive false alarms due to, for example, a high load condition. A sufficiently high alarm level results in reduced accuracy in the response of the ultrasonic monitoring system to specific machine fault conditions. An integrated solution with a predictive pattern recognition system that would predict the ultrasonic data level based on machine operational conditions is proposed as a solution to the high sensitivity of the ultrasonic monitoring system. The pattern recognition algorithm is able to create dynamic or variable alarm conditions which is offered to improve the accuracy of determining the health of the machine and/or the presence of mechanical faults based on the ultrasonic data analysis.