Piezoelectric-array-based MISO diagnostic system for CNN-condition monitoring of bearing/gearbox instruments

Abstract

The article presents a novel MISO (multi-input-single-output) diagnostic system suitable for spatial condition monitoring of bearing/gearbox instruments with multi-location defects. The sensor array consists of three piezoelectric patches: one is attached to the surface of the bearing house and the other two connected in parallel are mounted on the wall of the planetary gear. These two sets of patches are electrically connected in series for sensing the fault signals whose sources of anomalies come from either the bearing or the gear. They offer an advantage of allowing a single voltage output from multiple inputs. In addition, two inductances are connected to the sensor array to form LC resonant circuits for filtering the irrelevant noise at high frequency. A convolutional neural network (CNN) classifier is trained by 12x150 FFT spectrums. The result from the testing data with 12x10 FFT spectrums shows that the average accuracy is achieved to be as high as 92:5%, confirming the soundness of the proposed model.

Multi-location fault detection with piezoelectric arrays and multi-task CNN learning

“…To optimize the frequency response, it can be shown that the inductance magnitude L is determined by the resonant condition L * = 1 w 2 Cp . 38 Next, for spatial condition monitoring in rotational machines, a sensor array model developed by Lo et al 38 is briefly described here. In the present setup shown in Fig.…”

“…To differentiate between the 12 health states detailed in Table 1, Lo et al developed a deep learning approach using CNN as outlined in their work. 38 CNN efficiently extracts local features from input data through convolutional layers with various filters, reducing the neural network size through pooling while preserving essential information. The CNN architecture, depicted in Fig.…”

“…3 The mentioned CNN has demonstrated its capability in classifying various health states within a torque transmission device, covering rotating bearing and gear elements. 38 However, there are two drawbacks. Firstly, the number of categories to be classified is 12, resulting from the combination of 4 bearing defects and 3 gear defects.…”

“…It utilizes piezoelectric array sensors and multi-task CNN learning. The former employs a multi-input single-output (MISO) sensing approach by capturing signals from a piezoelectric patch array, [38][39][40][41][42][43] facilitating sensor fusion for spatial condition monitoring through a single voltage signal. The latter employs self-transfer learning to achieve defect classification with two output layers.…”