Abstract: This paper proposes an acoustic diagnosis technique for detecting abnormalities in and deterioration of machines that emit intermittent sounds during operation. The effectiveness of this technique is demonstrated experimentally. Acoustic diagnosis is generally applied to continuous sounds by analyzing the power spectrum patterns of regular, periodic sounds emitted by rotating components. However, machines such as automatic teller machines (ATMs) emit intermittent, episodic sounds during operation, making it impossible to employ the same diagnosis techniques as those used for conventional, continuous sounds. The proposed technique enables intermittent acoustic abnormalities to be diagnosed. It achieves this by constructing two vector series that are polygonal chain approximations of the temporal changes in the pressure levels of the most characteristic frequencies of the acoustic emissions during normal operation (the "standard vector series") and during inspection (the "measured vector series"). The technique employs dynamic programming (DP) matching to collate and compare the two vector series at standard intervals. The technique consists of the following six steps: (1) acquisition of the temporal changes in the pressure level, as acoustic data; (2) extraction of the diagnosis regions; (3) selection of relevant features using a polynomial expansion filter; (4) polygonal chain approximation of the acoustic waveforms by vector series; (5) collation of the resulting measured vector and standard vector series by DP matching; (6) diagnosis of abnormality by vector dissimilarity. This paper provides detailed descriptions of steps 3 to 6. Steps 3 and 5 are particularly notable: in step 3, the acoustic data are approximated as vectors in a polygonal chain using a Hermite polynomial and the relevant features are extracted; in step 5, the DP collation absorbs operational asynchronicities, thereby eliminating what has been the greatest impediment to intermittent sound diagnosis. The effectiveness of this method for localizing and diagnosing abnormalities is demonstrated experimentally by applying it to acoustic data from the paper-slip transport in an actual machine.

Brief Biography of the Speaker:
Teruji Sekozawa is a professor of Information Systems Creation at Kanagawa University, Japan. His area of expertise is the social information system, automotive control system, information service system. He authored or co-authored over 100 scientific papers published in reviewed journals or presented at conferences. He has over 150 patents related to system engineering. He had been researching the system development of the new social infrastructure in Hitachi Ltd. until 2004, and works in present Kanagawa University, Japan since 2005. He experienced the Chairman of Industrial system in SICE(Society of Instrument and Control Engineers) and the director of Tokyo branch in IEEJ(Institute of Electrical Engineers of Japan) as Scientific Activities. He is a senior member of IEEJ.