Data acquisition and processing for tip timing and operational modal analysis of turbomachinery blades

Abstract

In power generation turbines and aerospace engines, turbine blades life is limited by low cycle fatigue and high cycle fatigue, as well as deterioration due to foreign objects as sand and water. In order to ascertain the health of a turbine can be used the Blade Tip-Timing (BTT), an on-line watching and monitoring system. The use of a monitoring system is justified by economical and safety reasons. Blade failure will cause an extremely high cost to repair or replace the turbine and an excessive downtime. Blade Tip Timing also proves to be useful in the design of turbines as it allows validating the theoretical models using data measured under operational conditions. In Tip Timing, the arrival time of each blade tip is measured by sensors mounted at the engine casing. From the variations of the arrival times, the blade vibrations can be computed. The blade eigenfrequencies and damping ratios in operational conditions can then be determined by applying Operational Modal Analysis to these blade vibration data. The fact that the sensors can be mounted on a static part of the engine constitutes a major advantage over the traditional blade vibration measurements that use strain gauges mounted at the root of the rotating blades. However, a major problem for spectral and modal analysis of blade tip-timing data is that the signals are undersampled and aliased (Figure 1: Run-up of a blade with Aliasing (described as folded lines).). Indeed, the sampling frequency cannot be chosen in relation to the bandwidth of the vibration signals, but is determined by the number of sensors positioned around the casing circumferential and the engine rotation speed. This leads in general to undersampling. (Figure Presented) The method proposed in this paper is illustrated by means of measurements performed on power plant turbines using a set of inductive sensors mounted on the casing and tacho sensors mounted directly on the shaft. In order to increase the number of samples per revolution, the measurement was done using three sensors, all positioned on the same normal plane, with respect to the shaft direction, and at the same radial distance. Ideally, the sensors yield a sharp peak when a blade is passing, which makes determination of the times of arrival a straightforward process. In considered application the tip passage gives a relatively slow growth of the amplitude, compared with the ideal step signal; this involves the need of a careful selection of the threshold level for the detection of the passage timing. The dimension of the considered machines implies that the sensors have to be connected with long cables to the data acquisition front-end. The combination of the inductance of the sensor with the capacitance of the cable creates resonances into the data. Moreover the use of a magnetic sensor means that the amplitude of the signal is function of the material of the blade and of its manufacturing process, that could be the cause of residual magnetic charge; the amplitude is also function of the distance between the sensor and the tip of the blade and of the speed of passage of the tip. That entails the need of a threshold level that is function of the individual blade and of the RPM. This paper will discuss the following data acquisition and processing steps for performing Operational Modal Analysis (OMA) from turbomachinery Blade Tip Timing data: ▶ The inclusion of an additional resistor in the measurement chain that damps the electrical resonance with the aim to improve the arrival pulse data quality; ▶ A signal processing approach to detect arrival times, based on high-quality 24-bit sensor data acquisition; ▶ An innovative and accurate method to convert the data from fixed sensor position (i.e. fixed angle) to equidistant time domain data; ▶ Computation of the blade displacements from Blade Tip Timing data; ▶ Application of Operational Modal Analysis to the displacement data of each blade.(Figure 2: Operational Modal Analysis with LMS Test.Lab); ▶ Analysis of the aliasing from non-stationary data. (Figure Presented). © 2012 American Institute of Physics.