Search

This article presents machinery fault diagnosis in electric motors through mechanical vibration monitoring. Using a commercial machinery fault simulator, two fiber Bragg grating (FBG) accelerometers are evaluated under one healthy condition and nine different fault conditions, including rotors with broken bars, faulted bearings, misalignment, and imbalance. The mechanical vibration spectrum for each fault condition is assessed by analyzing the Bragg wavelength shift modulated by the acceleration. Offline analysis is carried out to identify the mechanical vibration pattern characteristics. In this preliminary analysis, the high-order frequency characteristics of faulted-bearing conditions and the slip frequency characteristics of broken rotor bar conditions were well-identified by both accelerometers. Following this analysis, this article uses two concurrent approaches to indicate the feasibility of classifying and/or clustering the data from supervised or unsupervised approaches. To that extent, the sensor system is first analyzed using a supervised approach from the $k$ -nearest neighbors (kNN) approach for fault condition classification, where the proposed supervised algorithm resulted in an accuracy of 100% for the fault condition estimation. After the classification analysis, data clustering is also proposed. In the unsupervised approach, the highest score for seven clusters was observed. Besides in this work, nine fault conditions were studied, there are two fault conditions where the positions of the accelerometers were interchanged for a sensor’s acceleration sensitivity evaluation purpose (maintaining the same fault condition). Finally, the results obtained in this work have shown the feasibility of the use of FBG-based accelerometers in different fault conditions under supervised and unsupervised approaches.

This article presents the development and degradation analysis of a liquid crystal (LC) sensor using a hydroxypropyl cellulose (HPC)–water mixture to achieve the Bragg effect in visible wavelength region. In this case, the LC can be used as a sensor for force monitoring. The tests are performed by means of applying controlled forces at the center of the sensor, where the red, green, and blue (RGB) components are acquired from computer vision algorithms. The results indicated the correlation between the RGB components and the applied force, where it is possible to infer the applied force from the RGB components using a multifeature linear regression. Furthermore, the degradation of the sensor is evaluated from the tests along 19 days using the same protocols in all tests, which indicated a variation on the sensor response as a function of the degradation. Another analysis indicated that the LC sensor degradation can be estimated from the RGB components in the border of the sensor region, which resulted in the possibility of compensating this effect using the degradation state as one of the inputs of the force estimation model. Results show that the proposed compensation technique can lead to force errors as small as 0.3 N.

This article presents the development and analysis of a hydroxypropyl cellulose (HPC)-based sensor for force estimation. The sensor is based on the mixture between the HPC and deionized water with different concentrations. The red-green-blue (RGB) components of the sensors responses are analyzed as a function of the concentration of HPC, which indicate the relation between the stable color and the concentration. In addition, the experiments with the force variation on the sensor system indicate the correlation between the concentration of the HPC and the sensor performance, where the sample with 63% concentration (in weight) demonstrated a higher sensitivity for red and green components. It is also worth noting that there is the possibility of measuring the force distribution along the HPC sample, where the effect of camera illumination is also analyzed, where an increase on sensor sensitivity is obtained in all analyzed cases. Furthermore, the analysis of sensitivity variation along the HPC sensor is performed by applying forces at different positions on the HPC sensor, where it is possible to observe a higher uniformity on the sample with 57% concentration. Such sample is used on the 2-D shape reconstruction of the device for measuring the force distribution along the sample, which demonstrated the feasibility of the proposed device on force distribution assessment with sub-centimeter spatial resolution. Finally, the use of a convolutional neural network (CNN) for image processing is implemented to increase the accuracy of the proposed device, which resulted in an average mean squared error (mse) of 0.037.

This article presents the fatigue analysis on a diaphragm-embedded optical fiber sensor after more than 40000 stress–strain cycles. The optical fiber sensor is based on a fiber Bragg grating (FBG) embedded in a rubber diaphragm through a vulcanization method. To evaluate the influence of the temperature variation in the sensor system, the diaphragm-embedded FBG is characterized as a function of the temperature, where a linear response (determination coefficient higher than 0.99) is obtained. To compensate the temperature influence on the sensor response, a bare FBG is added. Furthermore, the diaphragm-embedded FBG is subjected to the transverse displacement cycles and the wavelength shift is analyzed as a function of time (and number of cycles). Results show that the sensor performance is affected by the number of cycles, related to the fatigue on both diaphragm and optical fiber materials. A 10% reduction on the sensor sensitivity is obtained, where such reduction follows an exponential trend with sharper reduction in the middle of the test, followed by a saturation trend with higher number of cycles. Considering the sensitivity, the experimental analysis indicated a necessity to perform additional calibrations on the sensor system after 5000 cycles, after the middle region of the test. In addition, as the rubber is a viscoelastic material, the number of cycles also resulted in variations in the dynamic responses of the sensor. The decay of the sensor responses and the delay between strain and stress responses resulted in variations in 22.2% and 3.9%, respectively.