Convolutional neural networks (CNNs) with log-mel audio representation and CNN-based end-to-end learning have both been used for environmental event sound recognition (ESC). However, log-mel features can be complemented by features learned from the raw audio waveform with an effective fusion method. In this paper, we first propose a novel stacked CNN model with multiple convolutional layers of decreasing filter sizes to improve the performance of CNN models with either log-mel feature input or raw waveform input. These two models are then combined using the Dempster–Shafer (DS) evidence theory to build the ensemble DS-CNN model for ESC. Our experiments over three public datasets showed that our method could achieve much higher performance in environmental sound recognition than other CNN models with the same types of input features. This is achieved by exploiting the complementarity of the model based on log-mel feature input and the model based on learning features directly from raw waveforms.

Equipment condition monitoring and diagnosis is an important means to detect and eliminate mechanical faults in real time, thereby ensuring safe and reliable operation of equipment. This traditional method uses contact measurement vibration signals to perform fault diagnosis. However, a special environment of high temperature and high corrosion in the industrial field exists. Industrial needs cannot be met through measurement. Mechanical equipment with complex working conditions has various types of faults and different fault characterizations. The sound signal of the microphone non-contact measuring device can effectively adapt to the complex environment and also reflect the operating state of the device. For the same workpiece, if it can simultaneously collect its vibration and sound signals, the two complement each other, which is beneficial for fault diagnosis. One of the limitations of the signal source and sensor is the difficulty in assessing the gear state under different working conditions. This study proposes a method based on improved evidence theory method (IDS theory), which uses convolutional neural network to combine vibration and sound signals to realize gear fault diagnosis. Experimental results show that our fusion method based on IDS theory obtains a more accurate and reliable diagnostic rate than the other fusion methods.

To ensure the safety and reliability of equipment and effectively prevent the occurrence of major accidents, a monitoring and diagnosis system of equipment condition is proposed in this paper. First, a perceptual model of four layers, which can collect the original data of equipment by sensors and analyze the real-time information with intelligent algorithms and display the decision making on the screen, is designed. Second, a method of condition monitoring and diagnosis based on evidence weight is proposed. The basic probability assignment of evidence is corrected by Mahalanobis distance and sensor weight. A threshold is introduced to select the appropriate fusion rule by comparing the relationship between threshold and conflict factor. In order to verify the effectiveness and practicability of the method, the improved fusion algorithm is applied to the monitoring and diagnosis of centrifugal pumps. Finally, a prototype system is implemented to illustrate the validity of the system in practice.

As the information measured by a single sensor cannot reflect the real situation of mechanical devices completely, a multi-sensor data fusion based on evidence theory is introduced. Evidence theory has the advantage of dealing with uncertain information. However, it produces unreasonable conclusions when the evidence conflicts. An improved fusion method is proposed to solve this problem. Basic probability assignment of evidence is corrected according to evidence and sensor weights, and an optimal fusion algorithm is selected by comparing an introduced threshold and a conflict factor. The effectiveness and practicability of the algorithm are tested by simulating the monitoring and diagnosis of rolling bearings. The result shows that the method has better robustness.