Accurate prediction of concrete compressive strength is imperative for investigating the in-situ concrete quality. To avoid destructive testing, developing reliable predictive models for concrete compressive strength using nondestructive tests (NDTs) is an active area of research. However, many of the developed models are dependent on calibration and/or concrete past history (e.g. mixture proportion, curing history, concrete mechanical properties, etc.), which reduces their utility for in-situ predictions.

This paper develops predictive models for concrete compressive strength that are independent of concrete past history. To this end, ultrasonic pulse velocity (UPV) and rebound hammer (RH) tests were performed on 84 concrete cylindrical samples. Next, compressive strengths were determined using destructive testing on these cylinders, and predictive models were developed using NDT results. Furthermore, to ensure generalizability to new data, all models were tested on independent data collected from six different research papers. The results support combined usage of UPV and RH in a quadratic polynomial model structure. Therefore, the final model was proposed based on combining models from a threefold cross-validation of the experimental data. This model predicted the independent data with very good accuracy. Finally, a concrete quality classification table using combined RH and UPV is proposed based on a variant of machine learning k-means clustering algorithm.