The measurement of mechanical properties of metallic materials at high strain
rates has been challenging, notwithstanding the application of steel for
intermediate and dynamic loading conditions. This is due to a lack of
sophisticated measuring tools which will require a very high-speed camera to
capture the stages of deformation, with little availability of recent machines
capable of testing at high strain rates when compared with testing at quasi-static strain rates.

The quasi-static testing procedure has been well-established with different
international standards. Still, the dynamic testing procedures are very limited as
they are being modified from the quasi-static testing. It is quite challenging to
characterize the dynamic fracture toughness owing to limitations in the existing
standards such as BS 7448-3:2005. With the effect of inertia during the
experiment of high strain rates, many oscillations are generated which masks the true path of the load-displacement curve.

The concern spans from significant oscillations encountered with the stress-strain curve, making it difficult to obtain the dynamic mechanical properties of the material. Hence, it is difficult to include dynamic properties in the design of
structures, and this results in catastrophic failure whenever the material fails
under dynamic loading, and thus safety is not satisfied. As dynamic deformation
occurs with limited plastic deformation, the material fails without warning like
showing significant necking before failure.

In this research, X65 steel material was investigated and characterized at quasistatic and dynamic conditions using several techniques like instrumented Charpy test, tensile testing (flat and round specimen), fracture toughness test, and drop weight test, which led to the proposed methods of determining high strain rates material properties. An EDM notched and fatigue pre-cracked Charpy-sized specimens were utilised in this investigation. Quasi-static fracture toughness testing was used to characterize the material properties at low strain rates, which were applied in the machine learning algorithm to predict the material’s fracture toughness.

The finite element analysis was utilised to support the investigation of the stress
and strain distribution in a single-edge notched bend (SENB) specimen at varying
loading rates to determine the effect of loading rates and crack driving force for
the dynamic fracture toughness measurement. ABAQUS was employed in
performing the FEM simulations. The ductile and damage model parameters
were determined from experimental data using the Johnson-Cook model.

Analytical solutions were also implied through the application of irreversible
thermodynamics of dislocation evolution to predict the stress-strain curve at an
elevated strain rate. Damage constants for FEM calculations utilising the
Johnson-Cook model and the undelaying plasticity theory to capture the impact
of the strain rate were both utilised. The thermal diffusivity method was applied
to characterize the material behaviour at high loading rates, as it is being affected by the change in temperature to undergo an adiabatic process at dynamic loading. The change in temperature at elevated loading rates was taken into consideration in dislocation density theory for the application of body-centered cubic materials.

Due to the difficulty in determining data from the VHS Instron machine on
dynamic fracture toughness, the low-blow Charpy test was implemented to
determine the varying strain rate properties to correlate with the simulated results. Results from the experimental results show that material strength is affected by rates of loading and increases with loading rates, whereas fracture toughness decreases with the loading rates.

Finally, the machine learning approach was considered to predict the stressstrain curve and fracture toughness data. The training sets were derived from
experimental data with certain features including the strain rate to train the model. The random forest and multilayer perceptron regressor algorithm were utilised in this work for its application with small data sets and to reduce overfitting. The results showed that it is promising to predict material properties from the machine learning algorithm to reduce the cost of material testing. However, this has a limitation from the available number of datasets, which need to be derived from experiments to increase the accuracy of the prediction of dynamic fracture toughness.