On The Effectiveness Of Dimensionality Reduction For Unsupervised Structural Health Monitoring Anomaly Detection