One-dimensional Signals Research Articles

Adaptive thresholding and coordinate attention-based tree-inspired network for aero-engine bearing health monitoring under strong noise

In engineering, due to strong noise interference, it is a challenging issue for decision-making of the neural network to accurately detect healthy conditions of aero-engine bearings. As such, a hierarchical health monitoring model, termed adaptive thresholding and coordinate attention-based tree-inspired network (ATCATN) is developed for the health monitoring of aero-engine bearings under strong background noise. The ATCATN integrates the advantages of the adaptive thresholding and coordinate attention-based network (ATCAN) as well as a tree structure-based method, and it is not merely a straightforward integration of two models. First, the ATCAN is constructed as the backbone network, where the coordinate attention module is developed to process one-dimensional vibration signals; meanwhile, the designed deep residual adaptive thresholding module is introduced to extract significant features from vibration signals with strong noise. Second, a two-layer tree-inspired decision layer, consisting of seed and leaf nodes, is developed to reconstruct the output layer of the ATCAN. Finally, the trained ATCATN is more consistent with maintenance cognition involving multilevel decision information by the progressive determination of fault locations and fault sizes. The experimental results and comparative analysis with seven advanced methods on two aero-engine bearing datasets show that the developed model has the following advantages: (a) model robustness under strong noise; (b) more accurate diagnosis result and (c) hierarchical diagnosis with fault locations and fault sizes. Based on the experimental results, the ATCATN model demonstrates high accuracy in progressively identifying fault locations and sizes of aero-engine bearings even under strong noise interference, which is beneficial for formulating maintenance strategies and aligns with the basic understanding of maintenance.

Study on vibration signals identification method for pipeline leakage detection based on deep learning technology

For natural gas pipeline leakage systems based on distributed feedback fiber laser vibration sensors (DFB-FL), a novel neural network model (mCNN-LFLBs) integrating multi-scale convolutional neural networks (mCNN) with local feature extraction blocks (LFLBs) is proposed in this study. To address the problem of feature loss in the traditional signal image transfer process, the one-dimensional original signal is directly used as the input of the model, effectively maintaining the time-domain correlation of the leaked signal. The time-domain correlation of the leakage signal is preserved. To address the problem that traditional pattern recognition algorithms have a weak ability to extract deep features and local features, mCNN is established by this model, which realizes the use of convolution kernels of different scales to extract features of different frequency bands in one-dimensional signals. Additionally, LFLBs are introduced, which enhances the local expression ability of the model, further improves the feature mining quality of the model, and better presents the deep features of leaked signals. Experimental results demonstrate that the mCNN-LFLBs model has a classification accuracy of up to 98.6%, with a testing response time of approximately 10 ms, and outstanding performance in terms of convergence speed. These results highlight the potential practical application of the mCNN-LFLBs model in natural gas leak detection.

Improved Fault Diagnosis of Roller Bearings Using an Equal-Angle Integer-Period Array Convolutional Neural Network

This article presents a technique to carry out fault classification using an equal-angle integer-period array convolutional neural network (EAIP-CNN) to process the electrostatic signal of working roller bearings. Firstly, electrostatic signals were collected using uniform angle sampling to ensure the angle intervals between two adjacent data points stayed the same and the signal length was fixed to a pre-determined number of rotation cycles. Then, this one-dimensional signal was transformed into a two-dimensional matrix, where the component of each row was the signal in one period, and the ordinate value of each row represented the corresponding rotation period. Therefore, the row and column indexes of the matrix had a specific meaning instead of simply splitting and stacking the data. Finally, the matrixes were utilized to train the CNN network and test the classification performance. The results show that the classification rate using this technique reaches 95.6%, which is higher than that of 2D CNNs without equal-angle integer-period arrays.

Communication signal detection based on high-order cumulants time-frequency analysis: on the application of deep learning YOLOV5 network

The detection of communication signals in heterogeneous electromagnetic environments currently relies primarily on a one-dimensional statistical feature threshold method. However, this approach is highly sensitive to dynamic changes in the environment, fluctuations in signal-to-noise ratios, and complex noise. To address these limitations, this paper proposes a novel time-frequency diagram based on high-order accumulation for signal detection. Traditional time-frequency diagrams suffer from poor noise suppression ability and unclear features. However, higher-order cumulants can effectively overcome these shortcomings. Currently, methods based on higher-order cumulants are typically limited to one-dimensional signals. Yet, two-dimensional time-frequency signal diagrams can represent a broader array of features. This paper employs higher-order accumulation to extract time-frequency features from the received signal, thereby transforming the conventional radio detection problem into an image recognition challenge. By merging the advantages of higher-order accumulations and time-frequency diagrams, we propose the use of higher-order accumulation time-frequency diagrams for signal detection. Extensive experimental simulations demonstrate that the proposed time-frequency diagram exhibits strong anti-noise performance and effectively suppresses frequency bias from multiple perspectives. The performance of the Higher-Order Cumulant-Time Frequency (HOC-TF) indicated lower Root Mean Square Error (RMSE) compared with the Short-Time Fourier Transform-Time Frequency (STFT-TF) and Wavelet Transform-Time Frequency (WT-TF). Additionally, compared to the STFT-TF and WT-TF methodologies, the novel time-frequency diagram introduced demonstrates superior stability using the Singular Value Decomposition (SVD) method. Moreover, by combining the new time-frequency diagram with the deep learning YOLOV5 network, signal detection and modulation identification of communication signals can be achieved.

A small sample bearing fault diagnosis method based on novel Zernike moment feature attention convolutional neural network

Bearings are one of the core components of rotating machine machinery. Monitoring their health status can ensure the safe and stable operation of rotating machine equipment. The limited nature of bearing fault samples makes it difficult to meet the demand for sufficient samples based on deep learning methods. Therefore, how to solve the problem of small- samples is the key to achieving intelligent fault diagnosis. In bearing failures based on vibration signals, the complex operating environment causes the vibration signals to inevitably mix with noise. The mixing of fault signature features and noise intensifies the strong spatial coupling of different types of fault features. In addition, diagnosing bearing failures under different loads is challenging because of the complex working conditions of bearings. Given the above problems, a small sample-bearing fault diagnosis method based on a high and low-frequency layered algorithm (HLFLA) and a novel Zernike moment feature attention convolutional neural network (ZMFA-CNN) is proposed. First, the proposed HLFLA converts one-dimensional time series signals into two-dimensional signals distributed rectangularly according to different frequency bands, and is used to simplify network feature screening, reduce the impact of noise, and retain adjacent signal constraint information. In addition, a new ZMFA-CNN is proposed to further extract multi-order moment features and attention weights, and can significantly improve the model generalization ability without increasing model parameters. At the same time, it is combined with FilterResponseNorm2d and thresholded linear unit to further improve model performance. Finally, sufficient experiments verified that the algorithm proposed in this paper can solve the above problems and has excellent transfer generalization ability and noise robustness. In addition, the experimental results of applying the algorithm proposed in this article to gas turbine main bearing fault diagnosis prove the reliability of the algorithm proposed in this article.

Multi-input mutual supervision network for single-pixel computational imaging

In this study, we propose a single-pixel computational imaging method based on a multi-input mutual supervision network (MIMSN). We input one-dimensional (1D) light intensity signals and two-dimensional (2D) random image signal into MIMSN, enabling the network to learn the correlation between the two signals and achieve information complementarity. The 2D signal provides spatial information to the reconstruction process, reducing the uncertainty of the reconstructed image. The mutual supervision of the reconstruction results for these two signals brings the reconstruction objective closer to the ground truth image. The 2D images generated by the MIMSN can be used as inputs for subsequent iterations, continuously merging prior information to ensure high-quality imaging at low sampling rates. The reconstruction network does not require pretraining, and 1D signals collected by a single-pixel detector serve as labels for the network, enabling high-quality image reconstruction in unfamiliar environments. Especially in scattering environments, it holds significant potential for applications.

Automatic and real-time tissue sensing for autonomous intestinal anastomosis using hybrid MLP-DC-CNN classifier-based optical coherence tomography.

Anastomosis is a common and critical part of reconstructive procedures within gastrointestinal, urologic, and gynecologic surgery. The use of autonomous surgical robots such as the smart tissue autonomous robot (STAR) system demonstrates an improved efficiency and consistency of the laparoscopic small bowel anastomosis over the current da Vinci surgical system. However, the STAR workflow requires auxiliary manual monitoring during the suturing procedure to avoid missed or wrong stitches. To eliminate this monitoring task from the operators, we integrated an optical coherence tomography (OCT) fiber sensor with the suture tool and developed an automatic tissue classification algorithm for detecting missed or wrong stitches in real time. The classification results were updated and sent to the control loop of STAR robot in real time. The suture tool was guided to approach the object by a dual-camera system. If the tissue inside the tool jaw was inconsistent with the desired suture pattern, a warning message would be generated. The proposed hybrid multilayer perceptron dual-channel convolutional neural network (MLP-DC-CNN) classification platform can automatically classify eight different abdominal tissue types that require different suture strategies for anastomosis. In MLP, numerous handcrafted features (∼1955) were utilized including optical properties and morphological features of one-dimensional (1D) OCT A-line signals. In DC-CNN, intensity-based features and depth-resolved tissues' attenuation coefficients were fully exploited. A decision fusion technique was applied to leverage the information collected from both classifiers to further increase the accuracy. The algorithm was evaluated on 69,773 testing A-line data. The results showed that our model can classify the 1D OCT signals of small bowels in real time with an accuracy of 90.06%, a precision of 88.34%, and a sensitivity of 87.29%, respectively. The refresh rate of the displayed A-line signals was set as 300 Hz, the maximum sensing depth of the fiber was 3.6 mm, and the running time of the image processing algorithm was ∼1.56 s for 1,024 A-lines. The proposed fully automated tissue sensing model outperformed the single classifier of CNN, MLP, or SVM with optimized architectures, showing the complementarity of different feature sets and network architectures in classifying intestinal OCT A-line signals. It can potentially reduce the manual involvement of robotic laparoscopic surgery, which is a crucial step towards a fully autonomous STAR system.

A number sense as an emergent property of the manipulating brain

The ability to understand and manipulate numbers and quantities emerges during childhood, but the mechanism through which humans acquire and develop this ability is still poorly understood. We explore this question through a model, assuming that the learner is able to pick up and place small objects from, and to, locations of its choosing, and will spontaneously engage in such undirected manipulation. We further assume that the learner’s visual system will monitor the changing arrangements of objects in the scene and will learn to predict the effects of each action by comparing perception with a supervisory signal from the motor system. We model perception using standard deep networks for feature extraction and classification. Our main finding is that, from learning the task of action prediction, an unexpected image representation emerges exhibiting regularities that foreshadow the perception and representation of numbers and quantity. These include distinct categories for zero and the first few natural numbers, a strict ordering of the numbers, and a one-dimensional signal that correlates with numerical quantity. As a result, our model acquires the ability to estimate numerosity, i.e. the number of objects in the scene, as well as subitization, i.e. the ability to recognize at a glance the exact number of objects in small scenes. Remarkably, subitization and numerosity estimation extrapolate to scenes containing many objects, far beyond the three objects used during training. We conclude that important aspects of a facility with numbers and quantities may be learned with supervision from a simple pre-training task. Our observations suggest that cross-modal learning is a powerful learning mechanism that may be harnessed in artificial intelligence.

ASG-HOMGAT: a high-order multi-head graph attention network with adaptive small graph structure for rolling bearing fault diagnosis

Traditional Euclidean spatial data processing is difficult to capture the inherent relationships of unstructured data such as bearing vibration signals. Representing vibration signals in graphical form helps to preserve their topological structure and temporal information. Secondly, most existing graph convolutional network methods are based on large graph structured data, which incurs certain memory overhead when aggregating high-order neighborhood node information and ignores important information between samples in the global graph structure. To address these issues, this paper proposes a high-order multi-head graph attention network based on an adaptive small graph structure (ASG-HOMGAT) for fault diagnosis of rolling bearings. Firstly, the adaptive preprocessing layer is used to adaptively denoise and compress the one-dimensional time-domain vibration signal, generating small rule graph data with topological structure. Then, these small graph structured data samples are input into a higher-order graph neural network, which aggregates features from multiple higher-order neighborhoods to achieve richer feature representations and fully explore the intrinsic correlation between samples. Finally, these features are aggregated into a reinforced representation of graph nodes through a multi head attention mechanism, and a SoftMax classifier is used for fault classification. The experimental results show that the ASG-HOMGAT method has better performance compared to mainstream graph neural network diagnostic models. The code and model will be released at: https://github.com/ding-ss/ASG-HOMGAT.

Comparative analysis of VEP signals discrimination methods based on time-frequency transformation and CNN-2D

The Visual Evoked Potential (VEP) examination is used to analyze the appropriate functioning of the optical pathways from the retina to the visual cortex. In hospitals, the diagnosis made by physicians is based mainly on reading the temporal trace and identifying the latency P100. However, after a considerable research effort, it has been confirmed that this method is subjective and relatively less reliable. In our work, we report different approaches to resolve the inadequacy of traditional classification, by studying the efficiency of VEP signal classification in a comparative approach using 3 models: Model A: STFT-CNN, Model B: CWT-CNN, and Model C: Wigner-Ville-CNN, therefore we evaluate in the same context the effectiveness of using a pre-trained 2D CNN structure. The time-frequency transformation allows us to generate two-dimensional data from one-dimensional signals to bring out the integrated features that are not valued in the temporal plot, and then exploit them for good discrimination between the two classes, in order to be able to use a CNN-2D classification architecture, taking into consideration the advantages offered by this architecture in terms of the involvement of the attribute extraction phase and its efficiency in classifying 2D data. The results provided by the different scenarios proved that the Wigner-Ville transformation combined with a pre-trained CNN architecture can be considered a good method in terms of different performance metrics, which demonstrates that it is a successful candidate for providing significant assistance to physicians in their analysis of VEP signals.

Compound fault diagnosis of rolling bearings with few-shot based on DCGAN-RepLKNet

To guarantee the security of personnel on-site, diagnosing the malfunction of mechanical apparatus is imperative. The accomplishments within the domain of fault diagnosis have been partly attributed to the advancements in deep learning technology, which excels in feature extraction through extensive datasets. However, it is difficult to collect sufficient data to train high-precision fault diagnosis models in practice. A novel method called deep convolution generative adversarial network (DCGAN)-RepLKNet is proposed to address the challenge of gathering enough data to train high-precision fault diagnosis models in practice. This technique involves transforming a one-dimensional time series vibration signal into a two-dimensional (2D) time-frequency map via wavelet transform technology. Subsequently, DCGAN expands the 2D time–frequency map samples produced. Finally, RepLKNet is used to classify the fault samples. The proposed method has been verified in the PU compound fault data set and bearing real damage data set. The results show that the accuracy of this method has been improved by 5.70%, 6.34%, 9.08%, and 16.35% compared to 2D-CNN under different sizes datasets in case 1, and by 24.5% compared to 2D-CNN in case 2.

Optical fiber vibration signal recognition based on an efficient multidimensional feature extraction network.

In the field of optical fiber vibration signal recognition, one-dimensional signals have few features. People often used the shallow layer of a one-dimensional convolutional neural network (1D-CNN), which results in fewer features being learned by the network, leading to a poor recognition rate. There are also many complex algorithms and data processing methods, which make the whole signal recognition process more complicated. Therefore, an optical vibration signal recognition method based on an efficient multidimensional feature extraction network was proposed. Based on ResNet-50, efficient channel attention (ECA) was used to improve image features extraction ability, and a long short-term memory (LSTM) network was used to enhance the extraction of temporal features. Three different vibration signals were collected using a phase-sensitive optical time-domain reflectometry (Φ-OTDR) optical fiber sensing system. Vibration signals were converted into 128×128 grayscale images, which have more effective vibration information. The experimental results show that the three types of signals can be recognized and classified effectively by the network, and the average recognition rate is 98.67%.

AI-based preeclampsia detection and prediction with electrocardiogram data.

More than 76,000 women die yearly from preeclampsia and hypertensive disorders of pregnancy. Early diagnosis and management of preeclampsia can improve outcomes for both mother and baby. In this study, we developed artificial intelligence models to detect and predict preeclampsia from electrocardiograms (ECGs) in point-of-care settings. Ten-second 12-lead ECG data was obtained from two large health care settings: University of Tennessee Health Science Center (UTHSC) and Atrium Health Wake Forest Baptist (AHWFB). UTHSC data was split into 80% training and 20% holdout data. The model used a modified ResNet convolutional neural network, taking one-dimensional raw ECG signals comprising 12 channels as an input, to predict risk of preeclampsia. Sub-analyses were performed to assess the predictive accuracy for preeclampsia prediction within 30, 60, or 90 days before diagnosis. The UTHSC cohort included 904 ECGs from 759 females (78.8% African American) with a mean ± sd age of 27.3 ± 5.0 years. The AHWFB cohort included 817 ECGs from 141 females (45.4 African American) with a mean ± sd age of 27.4 ± 5.9 years. The cross-validated ECG-AI model yielded an AUC (95% CI) of 0.85 (0.77-0.93) on UTHSC holdout data, and an AUC (95% CI) of 0.81 (0.77-0.84) on AHWFB data. The sub-analysis of different time windows before preeclampsia prediction resulted in AUCs (95% CI) of 0.92 (0.84-1.00), 0.89 (0.81-0.98) and 0.90 (0.81-0.98) when tested on ECGs 30 days, 60 days and 90 days, respectively, before diagnosis. When assessed on early onset preeclampsia (preeclampsia diagnosed at <34 weeks of pregnancy), the model's AUC (95% CI) was 0.98 (0.89-1.00). We conclude that preeclampsia can be identified with high accuracy via application of AI models to ECG data.

A new approach for heart disease detection using Motif transform-based CWT's time-frequency images with DenseNet deep transfer learning methods.

Electrocardiogram (ECG) signals are extensively utilized in the identification and assessment of diverse cardiac conditions, including congestive heart failure (CHF) and cardiac arrhythmias (ARR), which present potential hazards to human health. With the aim of facilitating disease diagnosis and assessment, advanced computer-aided systems are being developed to analyze ECG signals. This study proposes a state-of-the-art ECG data pattern recognition algorithm based on Continuous Wavelet Transform (CWT) as a novel signal preprocessing model. The Motif Transformation (MT) method was devised to diminish the drawbacks and limitations inherent in the CWT, such as the issue of boundary effects, limited localization in time and frequency, and overfitting conditions. This transformation technique facilitates the formation of diverse patterns (motifs) within the signals. The patterns (motifs) are constructed by comparing the amplitudes of each individual sample value in the ECG signals in terms of their largeness and smallness. In the subsequent stage, the obtained one-dimensional signals from the MT transformation were subjected to CWT to obtain scalogram images. In the last stage, the obtained scalogram images were subjected to classification using DenseNET deep transfer learning techniques. The combined approach of MT+CWT+DenseNET yielded an impressive success rate of 99.31 %.

IInception-CBAM-IBiGRU based fault diagnosis method for asynchronous motors

Aiming at the problems of insufficient extraction of asynchronous motor fault features by traditional deep learning algorithms and poor diagnosis of asynchronous motor faults in robust noise environments, this paper proposes an end-to-end fault diagnosis method for asynchronous motors based on IInception-CBAM-IBiGRU. The method first uses a signal-to-grayscale image conversion method to convert one-dimensional vibration signals into two-dimensional images and initially extracts shallow features through two-dimensional convolution; then the Improved Inception (IInception) module is used as a residual block to learning features at different scales with a residual structure, and extracts its important feature information through the Convolutional Block Attention Module (CBAM) to extract important feature information and adjust the weight parameters; then the feature information is input to the Improved Bi-directional Gate Recurrent Unit (IBiGRU) to extract its timing features further; finally, the fault identification is achieved by the SoftMax function. The primary hyperparameters in the model are optimized by the Weighted Mean Of Vectors Algorithm (INFO). The experimental results show that the method is effective in fault diagnosis of asynchronous motors, with an accuracy rate close to 100%, and can still maintain a high accuracy rate under the condition of low noise ratio, with good robustness and generalization ability.

A Deep Learning Fault Diagnose Method for Turbine Bearing: Digital Twin Mechanism

This study presents a digital twin (DT) based wind turbine bearing fault diagnosis approach to address the issues of insufficient fault sample size and inaccurate diagnosis. To assist in diagnosing bearing faults in wind turbines, a DT system was built. Bearing vibration signal enhancement processing, which is based on the Hilbert-Huang transform, is used to improve the data samples of vibration signals and decrease the noise in these signals. In order to diagnose bearing defects in wind turbines, a convolutional neural network model was trained and tested using data-enhanced samples. The experimental results showed that the suggested method is feasible and effective, increased the stability and accuracy of defect diagnosis in wind turbine bearings, and solved the problem of data augmentation in one-dimensional vibration signals.

Gearbox fault diagnosis method based on improved semi-supervised MTDL and GAF

Aiming at the problem that it takes a long time and high cost to obtain complete labeled data under intelligent fault diagnosis and unlabeled data is not used. This paper proposes an improved semi-supervised mean teacher deep learning (MTDL) and Gramian angle field (GAF) fusion diagnostic method. This method fully utilizes a small number of labeled samples and a large number of unlabeled samples to deeply mine invisible fault features and potential physical correlations. At the same time, it solves the problem of losing the inter-data correlation structure when one-dimensional time series signals are used as inputs for neural networks. The GAF-MTDL method uses consistency regularization and modifies the network structure in the mean teacher algorithm into a semi-supervised deep learning model enhanced by WideResNet. The experimental results show that the proposed GAF-MTDL method saves a lot of manual labeling costs, improves the recognition accuracy and generalization ability, and can achieve excellent prediction accuracy with very little labeled data. In the end, the accuracy of planetary gear fault identification reached 98.22% under the labeling rate of 20%, and the accuracy of fault identification reached 99.98% through the verification of the bearing data set of Case Western Reserve University. The value of this research is to bring an efficient and low-cost technology to the field of industrial intelligent fault diagnosis, which can significantly improve the accuracy of fault identification.

Neural Network-Based Investigation of Periodic Noise Reduction Methods for High-Resolution Infrared Line Scanning Images

In the realm of neural network-based noise reduction, conventional models predominantly address Gaussian and blur artifacts across entire images. However, they encounter notable challenges when directly applied to periodic noise characteristics of high-resolution infrared sequential imagery. The high resolution also complicates the construction of suitable datasets. Our study introduces an innovative strategy that transforms two-dimensional images into one-dimensional signals, eliminating the need for processing the full image. We have developed a simulated dataset that closely mirrors natural infrared line scanning images derived from the FLIR dataset. To address low-frequency periodic noise, we propose two neural-network-based denoising approaches. The first employs a neural network to deduce noise from the one-dimensional signal, while the second utilizes discrete Fourier transforms for noise prediction within the frequency domain. Our experimental results highlight the Restormer model’s exemplary performance in direct noise prediction, where denoised images attain a PSNR of around 41 and an SSIM close to 0.9 on simulated data, leaving minimal residual noise in the actual denoised images. Further, we investigate the influence of Fourier coefficients, as predicted by neural networks, on the denoising process in the second approach. Employing 12 frequency coefficients, the Restormer and NAFNet models both reach a PSNR near 34 and an SSIM of approximately 0.842.

Domain adaptation-based sparse time-frequency analysis and its application on seismic attenuation estimation

Time-frequency (TF) transform is a commonly used tool for geologic structure interpretation and attenuation estimation, mainly including the linear and nonlinear methods. The former consists of short-time Fourier transform, continuous wavelet transform, S-transform, etc., which can be efficiently implemented. However, they suffer from the Gabor uncertainty principle, which limits their TF readability. The latter can effectively enhance TF readability; nevertheless, it has several unavoidable drawbacks, such as low computational efficiency and parameter selection. For seismic sparse TF (STF) analysis (STFA), we develop a domain adaptation-based STFA (DASTFA) model. This model mainly contains two steps. First, we suggest an STF network (STFNet) for mapping a 1D seismic signal to a 2D STF spectrum, which is pretrained using synthetic traces and STF labels. Second, based on the pretrained STFNet, we design a DASTFA model for transferring to field data. Note that the synthetic traces are generated using well-log data and horizon at the study area, located in the Ordos Basin, China, which can reduce the gap between synthetic and field data. Afterward, we use synthetic traces with their STF labels for model pretraining in the first step and field data without STF labels for model transferring in the second step. Here, model transferring indicates transferring the pretrained model parameters based on field traces. Finally, to test the validity and effectiveness of DASTFA, we apply it to synthetic and field data. Moreover, we test the availability of seismic attenuation estimation by adopting the suggested model for 3D poststack field data volume, which achieved encouraging results.

A comparative analysis of damage identification methods based on multi-channel data fusion and convolutional neural network for frame structures

Vibration-based damage identification methods are crucial for structural health monitoring. In recent years, deep learning methods, such as convolutional neural networks (CNNs), have been widely used for damage identification due to their superior performance. To enhance the accuracy of damage identification for complex frame structures and effectively utilize data measured from multiple sensors, this study proposes and compares four different damage identification methods based on CNN and four different kinds of multi-channel data fusion approaches, i.e., data splicing, matrix reconstruction, data dimension elevation, and sub-model integration. Firstly, the original one-dimensional vibration signals of the frame structures are measured. Then, multi-channel data fusion is performed on the signals to construct datasets used as input to train the CNN model. Finally, the sensitive damage features are automatically extracted using the trained CNN models after parameter tuning, and the structural damage patterns are identified. The effectiveness of these damage identification methods is evaluated using the numerical model of IASC-ASCE Benchmark structure, the Qatar University Grandstand Simulator test, and a three-layer stainless steel frame structure test. The efficiency of damage identification methods using multi-channel data is compared to those using only single-channel data. Additionally, the impact of noise on damage identification accuracy is also analyzed. The results demonstrate that the proposed four different damage identification methods based on CNN and multi-channel data fusion, outperform the methods that solely rely on single-channel data. Furthermore, these methods significantly enhance the accuracy of damage identification.