In a method for monitoring hardware devices of the hardware devices network to detect failures, past logged jobs are analyzed to determine a pattern of past usage of hardware devices by users of the hardware devices network. A failure of a hardware device is inferred based on a deviation in usage of the hardware devices from the pattern of past usage. In some embodiments, a state model is constructed that assigns one of two or more available operational states to each hardware device of the hardware devices network based on a recent load of jobs logged to the hardware device and a stored history of logged jobs indicative of typical loading of the hardware devices. One or more failing hardware devices are identified based on the assigned operational states.