Collaborative Neuro-Fuzzy in Edge-AI for Industrial Fault Diagnosis With Sensing Data Fusion

Edge Artificial Intelligence (Edge-AI) enabled Industrial Internet-of-Things (IIoT) is an emerging paradigm for industrial big data acquisition and smart applications, the edge layer can collect multisource data through interfaces, protocols and system integration. However, the complex and time-varying manufacturing environment greatly affects the quality of sensing data, and further brings significant challenge in carrying out the efficient data fusion for smart IIoT applications, e.g., fault diagnosis. To address this issue, within the Edge-AI paradigm, through enabling the deep convolution neural network (CNN) with the ability to deal with the data uncertainty and noise, a deep convolutional neuro-fuzzy approach using collaborative structure, named neuro-fuzzy, is proposed for industrial fault diagnosis. Through introducing a concept of optimal fusion set to obtain the valid sensing data, the optimal fusion degree matrix is first obtained to quantify the fusion degree of sensing data. Subsequently, the matrix is input into the integrated fuzzy domain to transform the values to corresponding fuzzy values, and the fuzzy values are processed in the deep CNN to achieve the convoluting representation for high-dimensional feature extraction. In the defuzzy layer, the fuzzy values with extracted features are finally converted into crisp ones to be fed into the fully connected layer for executing the data fusion and classification. With the experiments conducted on the real-world CWRU dataset, the comparisons and analysis have been shown to demonstrate the superiority of investigations.