Abstract: The accelerating increase of response quantities, such as strain and acoustic emission signals in the vicinity of failure time, has been revealed as a precursor of catastrophic failure. This critical precursor has been widely validated as a valid way to predict failure time by the retrospective prediction of volcanic eruptions, landslides and laboratory experiments of rock failures. But the scatter of exponents in the critical power law singularity relationship that describes acceleration in precursory signals leads to a debate on the actual value of critical exponent and the uncertainty of failure time prediction. Consequently, the uncertainty resulting from the scatter of exponents is a key difficulty in using such methods for prediction of the failure time through the use of acceleration precursors. Thus understanding the underlying mechanisms for the magnitude and variation of critical power-law exponents becomes a central problem for understanding the process of failure and failure time prediction. This paper presents a multi-scale damage mechanic model describing the accelerating process of time dependent failure. Theoretic derivations and demonstrations of critical power law relationship are presented for two typical load process, i.e. brittle creep failure under constant nominal stress and the load process of linearly increasing the nominal stress with time. It is found that values of the singularity exponents have a relationship with the parameter that defines the nonlinear level of damage evolution rate depending on to the local true stress. The physical expressions of critical parameters are deduced and the physical meanings of these critical parameters are explained. It is declared that the observed variation of the critical power law exponents not only due to the fluctuation in the measurement data, but has its intrinsic physical controls. Then a method is suggested to predict the failure time when the critical singularity exponent is unknown. This proposed methodology is validated through granite creep failure experiments in laboratory and the challenges for practical applications are demonstrated.