Early detection of looseness in bolted flanges is crucial for safety in the oil and gas industries. Failure to do so could result in catastrophic accidents like leakage and explosions. One way to indirectly detect looseness is by analyzing vibration signals from structures connected by such joints. If the structure can be sufficiently vibrated, changes in its vibration parameters may indicate looseness as it reduces joint and structural stiffness. Modeling dynamic systems with autoregressive models is advantageous, as changes in their parameters are a reflection of structural damage. In this study, an integrated parametric model algorithm was developed to detect loosening in flanged pipes. By using Autoregressive model parameters as looseness feature vector, a loose indicator based on Mahalanobis distance can be utilized for detection. The proposed approach was validated using both stationary (AR) and non-stationary (TAR) models adapted for flanged pipe structure vibrations. The structure was excited with white noise (stationary state), and with a moving mass inside the pipe (non-stationary state) to make the identification method practical. The proposed method was able to identify flange looseness at early stages using an output-only method, as shown in the results.