Research On Fault Diagnosis And Life Prediction Method Based On The Multi-channel Data Fusion

The multi-channel sensor data collected in the manufacturing system contains rich information,which provides the basis for the equipment status monitoring,fault diagnosis and remaining useful life prediction.However,there are complex spatial-temporal correlations in the multi-channel data,and the coupling between different data channels is strong.The existing analysis methods only focus on the temporal relevance of data in a single channel.The correlation between multi-channel data is ignored,which leads to the loss of key information and the problem of curse of dimensionality.It is difficult to effectively analyze the multichannel data.The accuracy,real-time processing and computational efficiency cannot meet the actual industrial application needs.This paper studies the complex correlation of the multichannel data,establishes the fault diagnosis and life prediction model by using data fusion methods.The proposed approaches can effectively excavate the key feature hidden in the data,and realizes high-precision condition monitoring,fault diagnosis and remaining useful life prediction.The main research contents include the following four aspects:(1)A multi-channel data analysis method based on the multilinear subspace learning algorithm is proposed.The non-redundant fault discrimination information is extracted by using the Multilinear Principal Component Analysis algorithm,without destroying the raw structure of the multi-channel data.The complex correlation in the multi-channel data can be effectively described.The Support High-Order Tensor Machine classification model is established,and CANDECOMP PARAFAC decomposition is adopted to process tensor data to improve the recognition ability and learning speed.The Cuckoo Search algorithm is used to optimize the hyper-parameters of the model,to further improve the accuracy of fault pattern recognition.Finally,the effectiveness of the proposed method is verified by the simulation multi-channel data set and actual case study.(2)A multi-channel data analysis method based on improved Convolutional Neural Network is proposed.The Multilinear Principal Component Analysis is adopted as a preprocessing method to extract non redundant features from the data,which can reduce the dimension of the original data and the influence of redundant information on the model.The dimension reduced data is used as the input of Convolution Neural Network,and the nonlinear features in the data are extracted and fused through multi-layer convolution and pooling.The Sostmax function is used to identify various fault modes.The real-world case study shows that the proposed model has higher diagnostic accuracy and faster training speed than the classical Convolution Neural Network model.(3)A feature self-learning model of the multi-channel data by using deep learning methods is proposed.The feature extraction,fusion and fault diagnosis are integrated into a deep learning framework.The features are extracted from both the temporal and spatial scales in parallel from the raw multi-channel data.The multi-layer fully connected neural network is established to fuse and compress the extracted features,and the improved Softmax function is adopted for the fault diagnosis.Finally,the effectiveness of the proposed method is verified by both the simulation study and real-world case study.(4)A remaining useful life prediction model based on the multi-channel feature fusion is proposed.After data preprocessing,the Stack Autoencoder is used to extract the key features in the data.The different degradation stages are divided by the correlation coefficients of features in different periods.In the rapid degradation stage,the features are extracted from the timefrequency domain of the raw data through the deep learning model with multiple inputs.The problem of inconsistent data length in different channels can be solved.The multi-channel feature fusion model based on global average pooling is established,and the fitting of degradation process is realized.Finally,the effectiveness of the proposed method is verified by a rolling bearing case.This paper proposes fault diagnosis and life prediction model based on the multi-channel data fusion,which can effectively identify the failure mode,and accurately predict the remaining useful life of equipments.The research provides the important basis for optimizing the equipment operation and maintenance strategy,and has important theoretical research and engineering application value.