In structures or machine components, fatigue failure is very common. It is initiated by a small defect which leads it to a catastrophic failure. The material defects, inclusion, impurities and machine operation can always be vulnerable to crack initiation and hence fatigue cannot be avoided. In metallic structures, the thermal loads can also alter the material properties such as young’s modulus, tangent modulus, yield stress, and ultimate tensile strength, etc. Consequently, in the presence of increasing temperature, it can be inferred that the material might become soft near the vicinity of the crack tip, which can lead to increase the size of the plastic zone under the same mechanical loads. Therefore, it is very complicated to estimate the retardation or acceleration of fatigue crack propagation under thermo-mechanical loads. This research investigates the interdependencies of crack depth and crack location on the dynamic response of a non-prismatic cantilever beam under thermo-mechanical loads. Temperature can influence the stiffness of the structure, thus, the change in stiffness can lead to variation in frequency, damping and amplitude response. These variations are used as key parameters to quantify damage of Aluminum 2024 specimen under thermo-mechanical loads. Experiments are performed on non-prismatic cantilever beams at non-heating (room temperature) and elevated temperature, i.e., 50°C, 100°C, 150°C and 200°C. This study considers a non-prismatic cantilever beam having various initially seeded crack depth (0.5 mm to 2.5 mm) and crack of 0.5 mm with natural propagation under load located at various locations, i.e., 5%, 10% and 15% of the total length from fixed end, respectively. The analytical, numerical and experimental results for all configurations are found in good agreement. Using available experimental data, a novel tool is formulated for in-situ damage assessment in the metallic structures for the first time under thermo-mechanical loads. This tool can quantify and locate damage using the dynamic response and temperature including the diagnosis of subsurface cracking. It fits around 82% of available data for validation within 10% of prediction error against a small change in the response parameter. The obtained results demonstrate the possibility to diagnose the crack growth at any instant within the operational condition under thermo-mechanical loads.