Published a technical paper about anomaly sensing method using the engine combustion sound data. Title “Acoustic monitoring on generator using Time Series Analysis Technique based on Fluctuation-Dissipation Theorem.”

MHI Power Control Systems Co., Ltd. has engaged in developing and manufacturing control systems for thermal power plants for many years. By focusing on the sensing and the prediction of anomaly in the thermal power process, we have been contributing our expertise and technology in order to maintain a reliable operation for our customers.
Joined with Professor Honda, department of science and technology at Keio University, our Senior Project Manager, Mr. Yonekura has contributed the above subject technical paper to the Society of Instrument and Control engineers. This is one of the results from their joint study in acoustic diagnosis of generators, which explains a possible anomaly sensing of gas turbine generators by using the engine combustion sound data. Anomaly diagnostic varies from the traditional methods to the analytical method based on the least squares or statistical processing methods. While they have attempted visualizing its data by Fast Fourier Transform (FTT), AR, Likelihood Ratio and Wavelet Analysis, etc., a systematic superiority has been realized in the Time Series Analysis Technique Theory basis Fluctuation-Dissipation Theorem, which is advocated by an experimental mathematician, the late Mr. Okabe. Based on this theory, applicability in detecting anomaly was validated. The result of validation by using the engine sound data shows that there is conformity in existing analyses and a certain possibility is acknowledged.
*Mr. Okabe’s method is to predict a changing point in stationarity of sensor signals. Validation of stationarity is carried out by applying 19 nonlinear permittivity conversion and 180 examinations. If all of the above passes the validation, it is considered as a proof of stationarity. If it fails, it is considered as anomaly. No parameter, threshold, any particular expertise or specific knowledge is required for attaining a result. That is a distinctive feature of his method.
To make this method for a practical use in the future, it is required to build up a robust system with an improvement of real time property, in addition to the further validations. They have also confirmed that a lower sampling cycle forces some effect on the result.
As the study result shows, the sensing of stationarity and anomaly can be determined clearly by this method. There seems to be a potential expansion to the further applied anomaly sensing in the operation process for variety of machinery.
“Instrument and Control: September Issue” published by the Society of Instrument and Control Engineers (SICE)
Attachment: Contributed technical paper at SICE ANNUAL CONFERENCE 2013 held in Nagoya from September 14 ~ 17, 2012