A procedure to monitor crack growth in Aluminum lug joints subject to fatigue loading is developed. Sensitivity analysis is used to decide sensor importance and monitor crack growth rate. A new feature extraction technique based on Discrete Cosine Transformation (DCT) is developed to analyze complex sensor signals. Self-sensing piezoelectric sensors are surface mounted on Al 2024 T351 lug joint samples, 0.25 in. thickness. Samples with single crack site and multiple crack sites were used in this study and to initiate multiple crack sites, they were notched symmetrically near the shoulders and then tested under a fatigue load of 110lbs (0.49kN) to 1100lbs (4.9kN). Crack lengths were monitored over the entire life of the lug joint sample using a CCD camera. Active sensing was carried out at every crack length, when the machined was stopped. The piezoelectric actuator was excited with a chirp signal, swept between 1kHz to 500kHz, and sensor readings were collected at a sampling rate of 2Ms/s. Using three different sensor sensitivity algorithms, the sensor signals are analyzed and their efficiency in predicting crack growth rates and deciding sensor importance is studied. Sensor sensitivity is defined as the changes observed in sensor signals obtained from a damaged sample compared to healthy sample. The first two algorithms, ORCA and One-Class SVM's, are based on statistical techniques for outlier detection and the third algorithm, a new detection framework, is based on feature extraction using Discrete Cosine Transformation (DCT). The efficacy of these methods for damage characterization is presented.