Failures of machines often cause serious damages. If those failures can be predicted, users of machines can be free from such serious damages. Our Failure Sign Detective Solution (FSDS) detects faint signs of machine failures and prevent failures. Machine operating data in time domain such as vibrations, sounds or movements, are monitored by sensors and converted to multidimensional frequency spectrum data by FFT. Those spectrum data obtained for the first several days when the machine is operating in normal modes are mapped on a two-dimensional plane, so called “Reference Map (RM)”, by using a data mapping scheme. Daily operating data are plotted on RM in real-time based on the geometrical relationships in between the daily data and data on RM in multidimensional space. When the machine operates in normal states, daily data stays in normal data clusters. However, deterioration starts at a part of the machine, daily data moves out from the normal clusters. From the distances between normal data and normal clusters, Failure Risk Index (FRI) is defined and warning is issued when FRI reaches the threshold that is set in advance. Although lots of challenges to failure prediction using machine learning or deep learning technologies, most have failed because enough amount of training data can not be prepared because failures rarely happen and various kinds of failures would happen. For the moment, FSDS will be only one solution that works in real circumstances. The same scheme can be applied to human cases. Like FSDS detects faint signs of failures with machines, it can detect faint signs of anomalies of human from movements. By monitoring heart rates or breathing sounds, stress or early states of diseases could be detected. From behavior of senior people, early state of dementia could be found. Risk of car driving could be also evaluated.