Research On Fault Diagnosis Methods Of Service Robot Based On Multi-Source Heterogeneous Data Fusion

In recent years,with the rapid development of neural network theory and computer hardware technology,data-driven fault diagnosis method has received more and more attention,especially for the deep learning-based fault diagnosis method,which has gradually developed into one of the mainstream technologies.Service robots are highly coupled electromechanical systems and usually equipped with various heterogeneous sensors.It is difficult to accurately locate and track faults using a single type of sensor data.It is necessary to comprehensively evaluate the operational status of the robots with multi-source heterogeneous data to ensure diagnosis accuracy.In this paper,the deep learning method is used as a technical means,and the overall goal is to improve the safety and reliability of service robot operation.We carry out the research on service robot fault diagnosis method with multi-source heterogeneous data fusion.The specific innovative works are as follows:To address the issue of traditional fault diagnosis algorithms failing to effectively extract heterogeneous fault features,a channel-wise convolutional neural network with a feature augmentation layer is proposed.First,a channel convolutional mechanism is developed to enable the convolutional kernel to correspond to the input channel data,allowing the model to extract heterogeneous data features separately.Second,a feature augmentation layer is designed to adaptively adjust the importance level of each channel’s feature map by evaluating the informational content of sensor features in different channels,which makes model’s focus on fault-related features.Finally,the fully connected layers are used to fuse heterogeneous features and obtain final fault diagnosis.The experimental results demonstrate that the proposed method can efficiently extract and fuse heterogeneous sensor fault information for service robots,with an average fault diagnosis accuracy increase of 3.63%compared to the best-performing comparative method.To address the challenge of extracting spatial-temporal fault features from multiple sensors,a data relationship graph construction method based on prior knowledge is proposed on the basis of system mechanism analysis,and a spatial-temporal difference graph convolutional neural network is also designed for spatial-temporal feature extraction.Firstly,the mathematical model of the service robot system is established to characterize the intrinsic relationship of multi-sensor data,and the mathematical model is used to guide the construction of the robot data relationship graph.Then,a spatial-temporal difference graph convolutional neural network model is constructed,in which a difference layer is established to calculate the local difference characteristics of nodes for feature enhancement,and the spatial-temporal convolution module is introduced to jointly capture temporal correlation and spatial correlation.The experimental results show that the average fault diagnosis accuracy of the proposed method in the service robot fault diagnosis is improved by 3.13%compared with the optimal comparison method.To address the issue of imbalanced service robot fault diagnosis datasets,a fault sample generation method based on multi-generator generative adversarial network is proposed.First,different generators are built for each type of sensor data to address the problem of a single generator being unable to comprehensively learn the distribution characteristics of multi-source heterogeneous data.Then,the proposed spatial-temporal difference graph convolutional neural network is used as the discriminator to enhance the model’s feature extraction and classification capability.Finally,adversarial training is employed to improve the performance of sample generation and discrimination,and the generated fault samples are used to augment the fault diagnosis dataset.Experimental results demonstrate that,under three different imbalanced data scenarios(1:2,1:5,and 1:10),utilizing a multi-generator generative adversarial network to augment the samples,the average fault diagnosis accuracy of the spatial-temporal difference graph convolutional neural network is improved by 2.25%,1.56%,and 2.28%,respectively,compared to the original dataset.