Anomaly detection

Human intuition tends to categorize visible information into meaningful and meaningless, harmful and harmless. These intuition-based decisions often lead to missing the initial symptoms of major problems.

A seemingly meaningless stream of data, such as the current flowing in an electronic circuit or the information flowing in a network, contains a large amount of information. Even if the stream data looks the same to human, it is actually a mixture of abnormal and normal, and meaningful and meaningless.

In this research, we are developing a waveform analysis tool @blewm (pronounced, abloom) by combining component analysis, data expansion, reinforcement learning, and spectral clustering. This research is also in the field of small-data fine-grained analysis technology.

Waveform analysis tool, @blewm, can be used for non-contact analysis of electronic devices. For example, it is possible to distinguish between normal and abnormal states without disassembling or processing the electronic equipment (Figure 3). It is also possible to carefully identify the points to be analyzed in the noise contained in the current of the outlet where multiple pulses are superimposed (overlapping) (Figure 4).

Part of this research is currently being conducted with the support of the New Energy and Industrial Technology Development Organization (NEDO).

Figure
Figure 3
Figure
Figure 4