This paper presents an approach to significantly improve modeling accuracy for the power and energy demands of industrial robots. This is achieved by taking the temperature dependency of the joint's viscuous friction parameters into account. While the connection is commonly known, it is usually neglected in state-of-the-art energy consumption models for industrial robots. This paper shows that a consideration of temperature-dependent friction provides significant improvement of energy modeling accuracy. The approach is validated on a test rig with a KUKA KR 16 robotic manipulator. Measurements show that the grid energy consumption modeling error can be reduced from up to 45% to approx. 5% over the whole spectrum of operating temperatures.