Study on uncertainty quantification and information gain strategy for non-destructive testing with incomplete measurement points

Non-destructive testing (NDT) of wood is an important technique for structural performance assessment. However, most wood NDTs face serious uncertainty problems when the measurement points are incomplete. Based on this, the motivation of this paper is to establish an uncertainty quantification method and its information gain strategy for wood NDT with incomplete measurement points under the Bayesian framework. Firstly, a Bayesian uncertainty model based on hyper-robust estimation is established, and two information gain strategies—the chord-mean method and the diameter-mean method, are proposed. Secondly, NDT based on impedance meter and stress wave is conducted on laboratory material samples, laboratory component samples, a thousand-year-old building and a centuries-old building, respectively, to establish the dataset. The quantification results are subsequently verified by the test set. Finally, the influences of different angles and the number of new measurement points on the information gain of the two methods are discussed. Moreover, the proposed method was applied to an engineering case combined with the stochastic finite element calculations based on Latin Hypercube sampling. The influence of the damage distribution results is verified through the on-site vibration test. The results show that the agreement rate between the proposed method and measurements is 86.86%. In addition, under the same number of measurement points, the information gain capability of the diameter-mean method is better than that of the chord-mean method, and the adding measurement paths at 45° and 135° can improve the results. Using the calibrated damage distribution, the maximum error of the modal parameters between the measured and calculated results is 4.12%. The research can significantly improve the field detection efficiency of large-scale or complex wood structures, and estimate the uncertainty, facilitating high-fidelity calculation models.