This paper proposes a method for automatic tuning of analysis parameter in the period analysis method ARS. Feature that indicates abnormalities in industrial machinery sometimes appear in relatively low-speed vibrations. ARS can detect these features because ARS has high resolution in the low frequency band. However, its parameters are currently determined heuristically. This is one of the challenges in applying ARS to practical systems. In this study, we propose to automatically adjust the parameters by taking advantage of the problems that exist in the ARS. Simulation results show that the parameters that were previously determined empirically can now be determined mechanically.

The Chain systems are a means of power transmission and are used for a variety of purposes. Chain elongation is a typical deterioration phenomenon in chain systems, and understanding and managing the elongation is extremely important to maintain system performance. This report explains the structure of chain elongation and introduces three methods for mechanically detecting elongation.

Owing to aging of distribution water pipe, the accidents such as leakage and burst are occurring. To prevent above accidents, it is essential to maintain and manage the system through appropriate renewal. For efficient maintenance and management, it is important to evaluate the condition of main by Non-Destructive Testing (NDT) and reflect it in renewal plans. One of method to maintain the main efficiently, deterioration diagnosis method for distribution water pipe focused on the eigen frequency change of in-plane bending mode in circular cylindrical shell is already suggested. The eigen frequency change of in-plane bending mode is expected to have high accuracy in detection of deterioration, because the eigen frequency is proportional to pipe thickness. However, the previous study was lack of the consideration of the implementation method for sensing equipment. In this study, we develop a vibration sensing actuation device that solves these problems. The device constructed from an IoT sensor module, an electrical relay circuit, a white noise generator and an electrodynamic exciter. The applicability of the device was evaluated using experimental circular ring to see if pipe thinning could be diagnosed. As a result, it was implied that the probability density of acceleration changed depending on the presence or absence of pipe thickness thinning.

In recent years, data analysis using AutoML has become popular. Since engineers in general are not necessarily familiar with machine learning for diagnosis of abnormalities and preventive maintenance of machines and structures, data analysis using AutoML has the potential to accelerate the introduction of machine learning technology in this field. While the machine learning platform DataRobot can implement algorithms with no code, it is necessary for humans to evaluate the balance between time cost and accuracy of the derived algorithms. In this study, we investigated the relationship between accuracy and processing time for each algorithm on binary classification, multinomial classification and regression, with the aim of optimizing the time cost and accuracy derived by DataRobot.

In this research, the system identification based on the maximum likelihood estimation method using the analytical solution of Fokker-Plank equation is discussed in case of an electrical LCR circuit system which is subjected to white noise excitation. In previous work, an author was already reported about the verification result of mechanical mass-spring system. However, there is a problem which the previous study didn’t consider the electrical circuit system. In order to expand the application area in our proposed identification method, it is necessary the applicability consideration to electrical circuit system. First, the expansion of the identification algorithm to an electrical LCR circuit system was conducted based on the current measurement. Furthermore, the numerical identification verification was conducted.

Noise countermeasures are an important engineering problem, and acoustic analysis methods for noise prediction and reduction have been widely studied. In a thermoacoustic engine, the sound is generated when the air vibrates due to the temperature gradient inside the pipe. Applying this effect may reduce noise by using the temperature distribution. Therefore, the purpose of this study is to analysis an acoustic space with a temperature gradient using a concentrated mass model. In this report, we consider the silencing effect of thermoacoustic silencer with open-end by conducting experiment and numerical calculation.

Sound and vibration are widely used in the diagnosis of rotating machinery failures. However, one of the challenges is the increasing amount of measured data. Therefore, we are building a method to identify spectra at measurement frequencies other then the Nyquist frequency using compressed sensing. By improving the conventional method and combining order analysis using rotation measurement, we have succeeded in reducing the number of measurement points and improving performance of compressed sensing, and this paper will report on the evaluation results.

Since rotating machineries are used and play crucial roles in various industries, it is important to diagnose condition of the machineries appropriately based on the knowledge of abnormal condition of the equipment. However, it is quite difficult to obtain the knowledge of abnormal condition. For this reason, simulation is expected to be an effective approach to estimate conditions of equipment instead of data acquisition. In this study, we focus on a diagnostic system that combines simulation and Artificial Intelligence. Simulating behavior of the rotating machinery including abnormal conditions in advance and using the result as training data, we construct a learning model that predicts the parameters of the simulation model from the measurement data. We consider a rotor shaft system and verify this method by predicting the weight as the amount of rotor unbalance through the learning model from the experimental measurement data.

This paper proposes a configuration to detect slight frequency changes in high-frequency bands by low-speed sampling, utilizing the accumulation for real-time serial-to-parallel converter (ARS). Recently, there has been significant interest in predictive maintenance through vibration monitoring. To obtain accurate waveforms in vibration measurement, we must set the sampling frequency to meet the sampling theorem. However, this requirement results in greater costs for memory and data traffic as the data volume increases. This is an obstacle to achieving low-cost and low-power consumption. A possible solution to this problem is to measure at a lower sampling frequency, although measurements with low-speed sampling cause aliasing. Aliasing is known to cause high-frequency signals to appear in low-frequency bands, and it is possible to calculate the frequency of the original signal by analyzing the signal appearing in low-frequency bands. However, the fast Fourier transform (FFT), which has conventionally been used as a typical method, suffers from the low-frequency resolution in low-frequency bands, so it is not suitable for analyzing aliasing signals. As a solution to this problem, we apply the ARS which is characterized by its high resolution in low-frequency bands and low computational complexity, to the aliasing signals. To verify the performance of the proposed method, we conducted a computer simulation. In this simulation, the pseudo-signal of a bearing with a damaged outer race was analyzed to estimate the ball pass frequency of the outer race (BPFO). The results show that ARS is capable of estimating BPFO, while FFT could not estimate BPFO appearing at low-frequency bands. Based on these results, the configuration of the proposed method is expected to be useful for the construction of low-cost and low-power predictive maintenance systems.