In this paper, we construct a one-dimensional convolutional neural network (1DCNN), which directly takes as the input the vibration signal in the mechanical operation process. It can realize intelligent mechanical fault diagnosis and ensure the authenticity of signal samples. Moreover, due to the excellent interpretability of the 1DCNN, we can explain the feature extraction mechanism of convolution and the synergistic work ability of the convolution kernel by analyzing convolution kernels and their output results in the time-domain, frequency-domain. What’s more, we propose a novel network parameter-optimization method by matching the features of the convolution kernel with those of the original signal. A large number of experiments proved that, this optimization method improve the diagnostic accuracy and the operational efficiency greatly.

Traffic sign detection and recognition plays an important role in expert systems, such as traffic assistance driving systems and automatic driving systems. It instantly assists drivers or automatic driving systems in detecting and recognizing traffic signs effectively. In this paper, a novel approach for real-time traffic sign detection and recognition in a real traffic situation was proposed. First, the images of the road scene were converted to grayscale images, and then we filtered the grayscale images with simplified Gabor wavelets (SGW), where the parameters were optimized. The edges of the traffic signs were strengthened, which was helpful for the next stage of the process. Second, we extracted the region of interest using the maximally stable extremal regions algorithm and classified the superclass of traffic signs using the support vector machine (SVM). Finally, we used convolution neural networks with input by simplified Gabor feature maps, where the parameters were the same as the detection stage, to classify the traffic signs into their subclasses. The experimental results based on Chinese and German traffic sign databases showed that the proposed method obtained a comparable performance with the state-of-the-art method, and furthermore, the processing efficiency of the whole process of detection and classification was improved and met the real-time processing demands.

Previous 1D convolutional neural networks (1D CNN) models for vibration fault diagnosis have high computational complexity and poor interpretability, which cannot meet the higher requirements of model storage, computational efficiency and reliability for airborne and portable devices. Considering these challenges, an Explainable and Lightweight 1D CNN (ELCNN) model based on Square Global Average Pooling (S-GAP) and improved for vibration signals is proposed. The feature extraction and classification layers of 1D CNN are optimized to minimize the model parameters and computational complexity and improve interpretability while ensuring diagnostic accuracy. The model compresses the number of convolutional layers, removes unnecessary bias, activation function and pooling layer, and replaces the fully connected layer (FCL) with S-GAP. Improved 1D CNN models of different methods are evaluated and analyzed on public datasets of rolling bearings. Results show that the ELCNN improved for vibration signals is more lightweight, anti-noise and explainable than other models, and the diagnostic accuracy is further improved.

This paper constructs a novel network structure (SVD-1DCNN) based on singular value decomposition (SVD) and one-dimensional convolutional neural network (1DCNN), which takes the original signal as input to realize intelligent diagnosis of bearing faults. The output of the first convolution layer was also analyzed from the perspectives of time domain and time-frequency domain in the simulation experiment. Through qualitative analysis and quantitative analysis, it was found that the convolution kernel not only extracted the classification features of signals but also gradually highlighted the learned features in the network training process. Moreover, applying this network in fault diagnosis of bearing date provided by the Case Western Reserve University (CWRU) Bearing Data Center, it was found that the convolution kernel could also achieve the above operation. The novel network of this paper achieved a good classification effect on both the simulated signals and the measured signals.

Effective intelligent fault diagnosis of bearings is important for improving safety and reliability of machine. Benefiting from the training advantages, deep learning method can automatically and adaptively learn more abstract and high-level features without much priori knowledge. To realize representative features mining and automatic recognition of bearing health condition, a diagnostic model of stacked sparse denoising autoencoder (SSDAE) which combines sparse autoencoder (SAE) and denoising autoencoder (DAE) is proposed in this paper. The sparse criterion in SAE, corrupting operation in DAE and reasonable designing of the stack order of autoencoders help to mine essential information of the input and improve fault pattern classification robustness. In order to provide better input features for the constructed network, the raw non-stationary and nonlinear vibration signals are processed with ensemble empirical mode decomposition (EEMD) and multiscale permutation entropy (MPE). MPE features which are extracted based on both the selected characteristic frequency-related intrinsic mode function components (IMFs) and the raw signal, are used as low-level feature for the input of the proposed diagnostic model for health condition recognition and classification. Two experiments based on the Case Western Reserve University (CWRU) dataset and the measurement dataset from laboratory were conducted, and results demonstrate the effectiveness of the proposed method and highlight its excellent performance relative to existing methods.

Internal leakage in the hydraulic actuators is concerned in this article, which is caused by seal damage, resulting in the limited performance of the system. To study the issue, this article proposes a method based on time–frequency analysis for the detection in hydraulic actuators. First, the pressure signal after filtering of the actuator in one side is collected when the control valve is affected by sinusoidal-like inputs. Second, the time–frequency image of pressure signal in a period at different leakage levels is obtained after continuous wavelet transform. Third is the sum of pixels in the time–frequency image. It is shown that the feature pattern is established by the sum of pixels in the time–frequency image that internal leakage and its severity could be detected effectively. The proposed method required two baselines and premeasured the pressure signal at 11 leakage levels. Once the sum of pixels in the time–frequency image values, obtained from the time–frequency image by continuous wavelet transform based on wavelet Cmor1-1 in subsequent offline tests, are greater than the first baseline, a leakage alarm is triggered. Furthermore, a severe leakage alarm is triggered when the value is greater than the second baseline. Experimental tests show the accuracy of the proposed scheme at different mother wavelets, and it is done without knowing the model of actuator or leakage.