In this paper, we propose a Transformer-based encoder architecture integrated with an unsupervised denoising method to learn meaningful and sparse representations of vibration signals without the need for data transformation or pre-trained data. Existing Transformer models often require transformed data or extensive computational resources, limiting their practical adoption. We propose a simple yet competitive modification of the Transformer model, integrating a trainable noise reduction method specifically tailored for failure mode classification using vibration data directly in the time domain without converting them into other domains or images. Furthermore, we present the key architectural components and algorithms underlying our model, emphasizing interpretability and trustworthiness. Our model is trained and validated using two benchmark datasets: the IMS dataset (four failure modes) and the CWRU dataset (four and ten failure modes). Notably, our model performs competitively, especially when using an unbalanced test set and a lightweight architecture.
12 June 2022 in Journal of Asian Architecture and Building Engineering
Observing the vibration of the elevator ropes in an elevator shaft is a difficult task. Although several methods of observing the displacement using image processing or laser displacement meters and then estimating the vibration through an observer or a Kalman filter have been developed, it is difficult to install sensors because of the movement of the elevator car and the thinness of the rope as an observation target. Therefore, in this study, we focus on a pulley called a compensating sheave that does not change its lateral position as the car and ropes move. The horizontal vibration of the ropes causes the compensating sheave to vibrate in the vertical direction. A method of estimating the horizontal vibration from this vertical vibration using arc length calculations and a short-time fast Fourier transform is proposed. First, the vertical vibration of the compensating sheave is distributed to the various mode responses based on the predominant frequency of the vertical vibration. The mode-decomposed vertical vibration is then converted into a horizontal mode response using an arc length relational expression. Analysis of seismic response data confirms that the proposed method can estimate the 1st, 2nd, and 3rd horizontal mode responses from the vertical vibration.
In this study, a concept and design of a self-powered sensor that utilizes a triboelectric effect to evaluate the condition of tensile bolted joints was proposed. Based on the fact that the triboelectric charge yields electrostatic voltage induced by the separation of the contacting rough surfaces, the proposed sensor is a film-shaped triboelectric sensor made of inexpensive materials being installed between the objects to be fastened. The principle of the sensor is that it detects microscale relative motions between the contacting surfaces against an external vibratory load when the integrity of the fastened joint is compromised due to a decrease in the bolt’s fastening force. In this study, we designed and fabricated triboelectric sensor and tested it on a tensile bolted joint specimen subjected to inertial vibratory loading, and it was experimentally shown that the output voltage amplitude of the sensor increased as the bolt’s fastening force decreased. In addition, a modeling study was performed to explain the unexpected decrease in voltage amplitude observed at medium preloads, by combining the triboelectric and mechanical models with the experimental results of two different external circuit configurations. Estimation of the triboelectric charge density at the contacting surfaces was performed, which was found to be consistent with the contact mechanics model assumed. Finally, the calculation of the sensor output voltage based on the presented mechanical/triboelectric model was provided, confirming the validity of the modeling study.
A smart gear sensor system has been developed for the condition monitoring of gear. This system includes a smart gear—the operation gear and a monitoring antenna. The analysis of the return loss of the monitoring antenna magnetically coupled with the smart gear gives the health condition of the gear. This research considers the effects of the distance and phase fluctuations between two components on the magnetic resonant return loss. The impacts of phase fluctuations include both static and high-speed conditions. Two experimental rigs have been built for the two cases. The coupling distance and static phase fluctuation are conducted via the first experimental rig. The second experimental rig performs both the coupling distance and phase fluctuation effect simultaneously while the smart gear rotates at high speed. During each test, the monitoring antenna return loss is captured thanks to a network analyzer. Analysis of the return loss data demonstrates that both the coupling distance and the change of the phase angle at the static condition and high speed have influenced the resonant return loss of the monitoring antenna. These findings are meaningful to the authors for evaluating and improving the accuracy of this gear health monitoring technique.
1 June 2023 in Journal of Mechanical Science and Technology