Non-destructive evaluating the density and mechanical properties of ancient timber members based on machine learning approach

• Machine learning models were proposed based on combined parameters of NDT and colour parameters. • Density, MOE, MOR, and CSPG were predicted accurately using the proposed ML methods. • The proposed ML models outperformed the MLR in predicting the density and mechanical properties of ancient timber members. The objective of this study was to evaluate the physical and mechanical properties of ancient timber members by combining non-destructive testing (NDT), colour parameters, and machine learning (ML) approaches. In this study, 175 small clear specimens were obtained from seven ancient timber members of different ages and used to obtain the parameters of NDT, apparent colour, and physical and mechanical properties. Pearson’s correlation among the parameters was analysed first, and relevance vector machine (RVM) models were developed after optimising the kernel function using the genetic algorithm method, while multiple linear regression (MLR) models were established as a comparison method. The results indicate that the developed RVM models have better evaluation performance than the MLR models. In addition, the mean absolute relative error between the test and predicted values was less than 10%, and R 2 was higher than 0.98 for the physical and mechanical property parameters in the verification tests, indicating that the developed RVM models had good generalisation ability and accuracy. Finally, an evaluation method for on-site application was proposed, particularly the method of obtaining the colour parameters of timber members. It was concluded that the proposed combined NDT, colour parameters, and ML approach provide an effective and accurate tool for the non-destructive evaluation of the density and mechanical properties of ancient timber members.