On determining the ductile–brittle transition temperature