Wavelet packet transformation has been used widely in the damage identification and structural health monitoring communities. In particular, the use of wavelet packet node energy (WPNE) as damage sensitive features has attracted much research interest in more recent years. WPNE features tend to contain detailed information which can be highly sensitive to local damage or other forms of structural changes. However, most of the existing studies in the literature on using wavelet energy based features have been numerical and involved idealised assumptions such as perfect and identical excitations among different tests. This paper presents an experimental investigation into the viability of WPNE based techniques for detection and localisation of the structural changes in a real measurement environment. Vibration signals are acquired firstly from the test structures with different alterations to the structural states, realized mainly through the use of additional masses, and WPNE features are extracted. These features and the corresponding structural states form a dataset, from which supervised machine learning with neural network is carried out. The trained neural network is subsequently tested for its prediction capability. The experimental structures include a free-ended steel I beam, a flat beam with fixed ends and MX3D Bridge, the world’s first 3D-printed metal bridge. Different forms of excitation are involved for different test structures, including hammer impact and controlled heel drops and impacts from pedestrian footfall. Results indicate that the WPNE based neural network approach is capable of detecting and localising the structural changes in all tested structures. The accuracy is generally higher in a better controlled excitation situation, where structural changes at a level equivalent to incipient damage is detectable.