Noise-Boosted Convolutional Neural Network for Edge-Based Motor Fault Diagnosis With Limited Samples

Abstract

Convolutional neural networks (CNNs) have been widely applied in motor fault diagnosis. However, to obtain high recognition accuracy, massive training data are typically required and transmitted to the cloud/local server for training, which may suffer from security and privacy problems. In this study, a noise-boosted CNN (NBCNN) model is developed to achieve accelerated training and improved recognition accuracy with limited training samples. First, the NBCNN model with a noise-injection fully connected layer is established. Then, a strategy for noise selection and injection is proposed to obtain an optimal matching among the data, model, and noise. Finally, the optimal injected noise accelerates the convergence of model training and improves the accuracy of motor fault diagnosis. Compared with the conventional CNN without noise injection and the state-of-the-art models, the effectiveness and superiority of the proposed NBCNN model are validated by two benchmark datasets. In addition, the algorithm is deployed onto an edge device and the results show that the training speed of the developed NBCNN can reach nine times faster than the conventional CNN. The proposed method shows remarkable potential for distributed model training, federal learning, and real-time motor fault diagnosis.

Research on convolutional neural networks in motor fault diagnosis

“…Using a combination of bi-spectrum and CNN, Ma et al [8] investigated the current signals pertaining to motor bearing faults. A model known as the noise enhanced CNN (NBCNN) was developed by Chen et al [9] to effectively and precisely diagnose motor faults. An efficient convolutional neural network (ECNN) model was developed by An et al [10] to enable real-time fault diagnosis and dynamic control.…”

Novel classification algorithms inspired by firing rate stochastic resonance

“…x 0 exp(z 2 )erfc(z)dz (28) then the new approximation can be acquired from fitting the function ( 28) by a linear combination of ln(x + 1) and x − √ x 2 + b plus a constant. We remark that the choice of this combination is mainly due to two points.…”

“…The first point is due to the fact that G(x) → +∞ very quick as x → −∞, and as a result the activation function can vanish very quick as well. Then, the fitting problem is reduced to fit the function (28) for x > 0. The second point is due to the fact that both ln(x + 1) and x − √ x 2 + b are very gradient-friendly [55], with simple derivative computation and fast calculation speed.…”

“…In the past few decades, a growing number of papers have actually witnessed a striking fact in the field of artificial intelligence: injection of suitable amount of noise can effectively improve the learning performance of the relevant artificial neural network (ANN). For example, the noise injection was found useful in boosting the generalization ability in the same way as Tikhonov regularization [21], accelerating stochastic gradient search [22,23] and expectation maximization [24,25], speeding up video classification [26], assisting online sensitivity identification [27], boosting motor fault diagnosis [28] and softening the hard threshold activation [29]. Nevertheless, in almost all of the noise-related classification algorithms, the role of noise has not been maximally utilized.…”

Lightweight and intelligent model based on enhanced sparse filtering for rotating machine fault diagnosis

“…The algorithms deployed onto the edge nodes can realize a fast and real-time processing of the sensor signals and can generate a timely fault diagnostic result and maintenance decision. A combination of ML and IoT provides a promising path for intelligent manufacturing and maintenance (Chen et al, 2023).…”