1D Convolutional Neural Networks and Applications: A Survey

Serkan KiranyazOnur AvciOsama AbdeljaberTurker InceMoncef GabboujDaniel J. Inman

article2019MSSP2,727 citations

Presents a comprehensive guide to 1D Convolutional Neural Networks, detailing how their compact architecture enables real-time, low-cost signal processing and high-accuracy classification in biomedical diagnosis, structural health monitoring, and fault detection.

Listen

The article surveys the development and engineering uses of compact 1D convolutional neural networks, which were introduced to handle one-dimensional signals such as electrocardiograms, vibration recordings, and motor currents. Traditional deep 2D CNNs excel on images but demand large labeled datasets and heavy computation, making them impractical for many 1D tasks where data are scarce or real-time operation on modest hardware is required. The review therefore focuses on lightweight 1D CNN architectures that perform only scalar convolutions and pooling, allowing them to fuse feature extraction and classification within a single trainable model.

The authors trace the evolution from early biological-neuron models through 2D CNN milestones such as LeNet and AlexNet, then detail the forward- and back-propagation equations specific to 1D layers. They evaluate performance on four major domains using publicly available benchmark data: patient-specific arrhythmia detection in ECG signals, vibration-based damage localization in large-scale civil structures, bearing-fault identification from motor currents, and open-circuit fault detection in modular multilevel converters. Across these studies the compact 1D networks reached accuracies of 97–99 percent while running tens of times faster than real time on ordinary CPUs.

These results matter because they demonstrate that high-accuracy, low-latency monitoring is feasible without specialized GPUs or exhaustive labeled datasets. In structural health monitoring, for example, individual 1D CNNs attached to wireless sensors detected loosened bolts instantly with zero false alarms. In cardiac monitoring the same approach enabled early warning of arrhythmias in healthy individuals by synthesizing plausible abnormal beats for training. The low computational footprint also supports deployment on mobile or embedded devices where power and memory are limited.

The article concludes that further gains are likely from replacing the linear neuron model with heterogeneous operators, as recently explored in operational neural networks. It recommends continued public release of code and datasets, systematic comparison against emerging alternatives, and extension to additional 1D signal domains. The principal limitation noted is reliance on the classic linear neuron; results therefore remain bounded by the expressive power of that model until more flexible architectures mature.

Izmir University of Economics Qatar University Tampere University of Technology University of Michigan anomaly detection classification computer vision 1D Convolutional Neural Network Electrocardiogram Structural Health Monitoring

arXiv: 1905.03554

Paper: Gradient-based learning applied to document recognition, Yann LeCun et al. (1998). Provides the foundational gradient-based convolutional architecture and backpropagation framework that the survey adapts and translates into one-dimensional signal formulations.
Paper: Deep learning for time series classification: a review, Hassan Ismail Fawaz et al. (2018). Establishes a comprehensive empirical benchmark of deep neural networks for 1D time-series classification, providing critical comparative context for dedicated 1D CNN architectures.
Paper: An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling, Shaojie Bai et al. (2018). Demonstrates the foundational effectiveness and architectural principles of 1D temporal convolutions over recurrent networks for general sequential data modeling.
Paper: An Introduction to Convolutional Neural Networks, Keiron O'Shea et al. (2015). Introduces the standard building blocks, parameter sharing, and pooling operations of convolutional networks that are reduced to scalar operations in 1D CNNs.
Paper: EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces, Vernon J. Lawhern et al. (2016). Presents a compact, specialized 1D/spatial convolutional architecture for biological time-series signals that exemplifies the lightweight edge-deployment motivations reviewed in the survey.
Paper: Temporal Convolutional Networks for Action Segmentation and Detection, Colin Lea et al. (2016). Pioneers the use of hierarchical temporal 1D convolutions for sequence modeling, illustrating early successes in 1D signal segmentation.
Paper: Efficient Processing of Deep Neural Networks: A Tutorial and Survey, Vivienne Sze et al. (2017). Explains the energy, memory, and computational trade-offs that motivate adopting compact 1D scalar convolutions over resource-heavy multidimensional models on hardware.
Paper: Recent advances in convolutional neural networks, Jiuxiang Gu et al. (2015). Surveys the core architectural enhancements and optimization techniques in standard CNNs that serve as the baseline designs surveyed in 1D applications.

Paper: ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks, Qilong Wang et al. (2019). Directly applies fast 1D scalar convolutions across channel dimensions to create an ultra-lightweight channel attention mechanism for deep architectures.
Paper: A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects, Zewen Li et al. (2020). Broadens the survey's focus by synthesizing modern 1D, 2D, and multi-dimensional convolutional architectures into a unified comprehensive framework.
Paper: On the Relationship between Self-Attention and Convolutional Layers, Jean-Baptiste Cordonnier et al. (2020). Investigates the theoretical and operational equivalence between attention mechanisms and convolutional layers, offering theoretical depth to convolution design.
Paper: CvT: Introducing Convolutions to Vision Transformers, Haiping Wu et al. (2021). Extends the efficiency advantages of convolutional operators by embedding depthwise convolutions directly within modern transformer architectures.
Paper: Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention, Angelos Katharopoulos et al. (2020). Explores linear-complexity sequence modeling alternatives that address sequence processing efficiency alongside compact convolutional approaches.

Cover for 1D Convolutional Neural Networks and Applications: A Survey

Abstract

During the last decade, Convolutional Neural Networks (CNNs) have become the de facto standard for various Computer Vision and Machine Learning operations. CNNs are feed-forward Artificial Neural Networks (ANNs) with alternating convolutional and subsampling layers. Deep 2D CNNs with many hidden layers and millions of parameters have the ability to learn complex objects and patterns providing that they can be trained on a massive size visual database with ground-truth labels. With a proper training, this unique ability makes them the primary tool for various engineering applications for 2D signals such as images and video frames. Yet, this may not be a viable option in numerous applications over 1D signals especially when the training data is scarce or application-specific. To address this issue, 1D CNNs have recently been proposed and immediately achieved the state-of-the-art performance levels in several applications such as personalized biomedical data classification and early diagnosis, structural health monitoring, anomaly detection and identification in power electronics and motor-fault detection. Another major advantage is that a real-time and low-cost hardware implementation is feasible due to the simple and compact configuration of 1D CNNs that perform only 1D convolutions (scalar multiplications and additions). This paper presents a comprehensive review of the general architecture and principals of 1D CNNs along with their major engineering applications, especially focused on the recent progress in this field. Their state-of-the-art performance is highlighted concluding with their unique properties. The benchmark datasets and the principal 1D CNN software used in those applications are also publically shared in a dedicated website.

1 Introduction
2 Overview of Convolutional Neural Networks
2.1 2D Convolutional Neural Networks
2.2 1D Convolutional Neural Networks
2.3 Forward- and Back-Propagation in CNN-layers
3 Applications of 1D CNNs
3.1 Real-time Electrocardiogram (ECG) Monitoring
3.2 Vibration-Based Structural Damage Detection in Civil Infrastructure
3.3 Condition Monitoring in Rotating Mechanical/Aerospace Machine Parts
3.4 Fault Detection in Modular Multilevel Converters (MMC)
4 Computational Complexity Analysis of 1D-CNNs
5 Conclusions
References

Knowls

Knowl 1 — Compact 1D Convolutional Neural Network Architecture for Raw Sequential Signals
model/method
A compact 1D Convolutional Neural Network (1D CNN) operates directly on raw one-dimensional signals without requiring explicit 1D-to-2D signal transformation (such as spectrogram computation) or handcrafted feature engineering. The architecture integrates feature extraction and classification into a unified, end-to-end trainable model consisting of two main functional components:
1. Hidden 1D CNN Layers: Perform feature extraction through a cascade of 1D valid convolutions, non-linear activation functions, and 1D subsampling (pooling) operations. Filter kernels and feature maps are strictly 1D arrays, and operations are linear array convolutions.
2. MLP Layers: Fully connected Multi-Layer Perceptron layers that perform classification or regression on the scalar outputs produced by the final CNN layer.
To handle variable-length input signals without altering network topology, 1D CNNs employ an adaptive subsampling mechanism where the pooling factor of the final CNN layer is automatically adjusted so that the output dimension matches the fixed input dimensionality of the subsequent fully connected layer. Compact 1D CNN configurations typically use shallow depths (e.g., $\le 3$ CNN layers and $\le 50$ total hidden neurons), allowing efficient training and inference on standard multi-core CPUs or low-power embedded edge devices without specialized GPU hardware.
Knowl 2 — Mathematical Formulation of Forward Propagation in 1D CNN Layers
equation
In a 1D Convolutional Neural Network, the forward propagation at a hidden convolutional layer $l$ calculates the convolved input $x_k^l$ and the downsampled output feature map $s_k^l$ for the $k$ -th neuron ( $k \in \{1, \dots, N_l\}$ ) as:

$x_k^l = b_k^l + \sum_{i=1}^{N_{l-1}} \text{conv1D}(w_{ik}^{l-1}, s_i^{l-1})$

$y_k^l = f(x_k^l)$

$s_k^l = y_k^l \downarrow ss$

where:
- $N_l$ is the number of neurons in layer $l$ , and $N_{l-1}$ is the number of neurons in layer $l-1$ .
- $s_i^{l-1}$ is the 1D output feature map vector of the $i$ -th neuron in layer $l-1$ (with $s^0$ representing the raw input signal).
- $w_{ik}^{l-1}$ is the 1D filter kernel vector connecting the $i$ -th neuron of layer $l-1$ to the $k$ -th neuron of layer $l$ .
- $b_k^l$ is the scalar bias of the $k$ -th neuron in layer $l$ .
- $\text{conv1D}(\cdot, \cdot)$ denotes a valid 1D convolution without zero-padding, meaning the output vector $x_k^l$ has a dimension smaller than the input vector $s_i^{l-1}$ by $(K - 1)$ , where $K$ is the kernel length.
- $f(\cdot)$ is an element-wise non-linear activation function producing the intermediate feature map $y_k^l$ .
- $\downarrow ss$ denotes 1D downsampling (subsampling/pooling) by a scalar downsampling factor $ss$ .
Knowl 3 — Back-Propagation and Parameter Sensitivities in 1D CNN Layers
equation
Given a Mean Squared Error loss function for an input pattern $p$ with target vector $\mathbf{t}^p = [t_1^p, \dots, t_{N_L}^p]^T$ and network output vector $\mathbf{y}^L = [y_1^L, \dots, y_{N_L}^L]^T$ at output layer $L$ :

$E_p = \sum_{i=1}^{N_L} (y_i^L - t_i^p)^2$

The back-propagation (BP) equations compute the sensitivities with respect to network parameters and intermediate layer signals as follows:
1. Neuron Input Delta Error: For neuron $k$ at hidden CNN layer $l$ , the delta error vector $\Delta_k^l = \frac{\partial E}{\partial x_k^l}$ is obtained from the output sensitivity vector $\Delta s_k^l = \frac{\partial E}{\partial s_k^l}$ by:
$\Delta_k^l = \text{up}(\Delta s_k^l) \beta f'(x_k^l)$

where $\text{up}(\cdot)$ denotes zero-order upsampling by factor $ss$ , $\beta = (ss)^{-1}$ is the scaling factor, and $f'(x_k^l)$ is the derivative of the activation function evaluated at $x_k^l$ .
1. Inter-layer Sensitivity Propagation across CNN Layers: The sensitivity vector $\Delta s_k^l$ is back-propagated from the delta vectors $\Delta_i^{l+1}$ of the next layer $l+1$ via:
$\Delta s_k^l = \sum_{i=1}^{N_{l+1}} \text{conv1Dz}(\Delta_i^{l+1}, \text{rev}(w_{ki}^l))$

where $\text{rev}(\cdot)$ reverses the 1D filter kernel array, and $\text{conv1Dz}(\cdot, \cdot)$ denotes full 1D convolution with zero-padding.
1. Transition from MLP to CNN Layers: Between the first MLP layer ( $l+1$ ) and the last CNN layer ( $l$ ), scalar back-propagation computes:
$\Delta s_k^l = \sum_{i=1}^{N_{l+1}} \Delta_i^{l+1} w_{ki}^l$
1. Weight and Bias Gradients and Parameter Updates:
$\frac{\partial E}{\partial w_{ik}^l} = \text{conv1D}(s_k^l, \Delta_i^{l+1}) \quad \text{and} \quad \frac{\partial E}{\partial b_k^l} = \sum_n \Delta_k^l(n)$

$w_{ik}^{l-1}(t+1) = w_{ik}^{l-1}(t) - \varepsilon \frac{\partial E}{\partial w_{ik}^{l-1}}, \quad b_k^l(t+1) = b_k^l(t) - \varepsilon \frac{\partial E}{\partial b_k^l}$

where $\Delta_k^l(n)$ is the $n$ -th element of $\Delta_k^l$ , $\varepsilon > 0$ is the learning factor, and $t$ denotes the training iteration step.
Knowl 4 — Iterative Back-Propagation Training Algorithm for 1D CNNs
algorithm
The supervised training of a 1D CNN over a dataset of labeled 1D signal records proceeds by iteratively updating filter kernels and biases across all convolutional and fully connected layers using stochastic gradient descent.
Input: Training dataset of 1D signal vectors and corresponding target class labels, learning rate $\varepsilon$ , maximum iterations $I_{max}$ , layer configurations $\{N_l\}_{l=1}^L$ , kernel dimensions, subsampling factors
Output: Trained 1D CNN kernel weights $w$ and biases $b$ across all layers $l \in \{1, \dots, L\}
Initialize all kernel weights $w_{ik}^l$ and biases $b_k^l$ randomly from a uniform distribution $U(-0.1, 0.1)$
for iteration = 1 to $I_{max}$ do
    for each 1D signal sample $p$ in training dataset with target vector $\mathbf{t}^p$ do
        // Forward Propagation (FP)
        for layer $l = 1$ to $L$ do
            for neuron $k = 1$ to $N_l$ do
                if layer $l$ is a CNN layer then
                    Compute input vector $x_k^l = b_k^l + \sum_{i=1}^{N_{l-1}} \text{conv1D}(w_{ik}^{l-1}, s_i^{l-1})$
                    Compute activation $y_k^l = f(x_k^l)$
                    Compute subsampled output $s_k^l = y_k^l \downarrow ss$
                else
                    Compute scalar input $x_k^l = b_k^l + \sum_{i=1}^{N_{l-1}} w_{ik}^{l-1} y_i^{l-1}$
                    Compute scalar activation $y_k^l = f(x_k^l)$
                end if
            end for
        end for
        // Back-Propagation (BP)
        Compute output layer error: $\Delta_k^L = \frac{\partial E_p}{\partial x_k^L}$ for all $k \in \{1, \dots, N_L\}$
        for layer $l = L-1$ down to 1 do
            for neuron $k = 1$ to $N_l$ do
                if layer $l+1$ is an MLP layer and layer $l$ is an MLP layer then
                    Compute $\Delta_k^l = f'(x_k^l) \sum_{i=1}^{N_{l+1}} \Delta_i^{l+1} w_{ki}^l$
                else if layer $l+1$ is an MLP layer and layer $l$ is the final CNN layer then
                    Compute $\Delta s_k^l = \sum_{i=1}^{N_{l+1}} \Delta_i^{l+1} w_{ki}^l$
                    Compute $\Delta_k^l = \text{up}(\Delta s_k^l) \beta f'(x_k^l)$
                else
                    Compute $\Delta s_k^l = \sum_{i=1}^{N_{l+1}} \text{conv1Dz}(\Delta_i^{l+1}, \text{rev}(w_{ki}^l))$
                    Compute $\Delta_k^l = \text{up}(\Delta s_k^l) \beta f'(x_k^l)$
                end if
            end for
        end for
        // Post-Processing (PP) and Parameter Update
        for layer $l = 1$ to $L$ do
            for each connection $(i, k)$ and neuron $k$ in layer $l$ do
                if layer $l$ is a CNN layer then
                    Compute $\frac{\partial E}{\partial w_{ik}^{l-1}} = \text{conv1D}(s_i^{l-1}, \Delta_k^l)$
                    Compute $\frac{\partial E}{\partial b_k^l} = \sum_n \Delta_k^l(n)$
                else
                    Compute $\frac{\partial E}{\partial w_{ik}^{l-1}} = \Delta_k^l y_i^{l-1}$
                    Compute $\frac{\partial E}{\partial b_k^l} = \Delta_k^l$
                end if
                Update $w_{ik}^{l-1} \leftarrow w_{ik}^{l-1} - \varepsilon \frac{\partial E}{\partial w_{ik}^{l-1}}$
                Update $b_k^l \leftarrow b_k^l - \varepsilon \frac{\partial E}{\partial b_k^l}$
            end for
        end for
    end for
end for
Knowl 5 — Computational Complexity Analysis of Forward and Backward Propagation in 1D CNNs
theoretical result
The computational complexity of 1D CNNs across $L$ convolutional layers is characterized by the total counts of scalar multiplications and additions performed per forward propagation (FP) pass and back-propagation (BP) iteration.

Let $N_l$ denote the number of neurons in layer $l$ , $s_l^l$ denote the length of the output vector of a neuron at layer $l$ , $w_l^l$ denote the kernel length at layer $l$ , and $x_l^l$ denote the input vector length of a neuron at layer $l$ . Ignoring boundary effects and bias additions:
1. Forward Propagation Operations: $T_{FP}(mul) = \sum_{l=1}^L N_{l-1} N_l s_l^{l-1} (w_l^{l-1})^2$ $T_{FP}(add) = \sum_{l=1}^L N_{l-1} N_l s_l^{l-1}$
2. Back-Propagation Operations (First Convolution for Delta Error Propagation): $T_{BP}^1(mul) = \sum_{l=0}^{L-1} N_{l+1} N_l x_l^{l+1} (w_l^l)^2$ $T_{BP}^1(add) = \sum_{l=0}^{L-1} N_{l+1} N_l x_l^{l+1}$
3. Back-Propagation Operations (Second Convolution for Weight Sensitivities): Where $w_l^l = x_l^{l+1} - s_l^l$ : $T_{BP}^2(mul) = \sum_{l=0}^{L-1} N_{l+1} N_l w_l^l (x_l^{l+1})^2$ $T_{BP}^2(add) = \sum_{l=0}^{L-1} N_{l+1} N_l w_l^l$
4. Total Per-Iteration Complexity: $\text{Total Multiplications} = T_{FP}(mul) + T_{BP}^1(mul) + T_{BP}^2(mul)$ $\text{Total Additions} = T_{FP}(add) + T_{BP}^1(add) + T_{BP}^2(add)$
The total addition count is negligible compared to the total multiplication count, particularly when the kernel sizes $w_l^l$ are large. Both operation complexities scale linearly with the total number of inter-neuron connections ( $N_{l-1} N_l$ ) across consecutive layers.
Knowl 6 — Patient-Specific and Personalized Early ECG Arrhythmia Detection Using 1D CNNs
empirical result
Compact 1D CNNs applied to real-time Electrocardiogram (ECG) beat classification achieve state-of-the-art diagnostic performance under two operational paradigms:
1. Patient-Specific ECG Classification: A dedicated compact 1D CNN trained on a small amount of patient-specific data (the first 5 minutes of a record plus 200 common beats) classifies raw ECG beats into five categories: normal sinus beats (N), supraventricular ectopic beats (S), ventricular ectopic beats (V), fusion beats (F), and unclassifiable beats (Q). Evaluated on 83,648 beats from 44 records of the MIT/BIH Arrhythmia Database, the compact 1D CNN achieves:
- Ventricular Ectopic Beat (VEB) average detection accuracy: $99.0\%$
- Supraventricular Ectopic Beat (SVEB) average detection accuracy: $97.6\%$
1. Personalized Early Warning for Healthy Individuals: To detect early cardiac arrhythmias when patient-specific abnormal beats are not available during training (as in healthy individuals), an Arrhythmia Beat Synthesis (ABS) filter library models cardiac anomaly mechanisms to synthesize potential abnormal beats from the individual's normal beats. Training a compact 1D CNN on the individual's real normal beats and synthesized abnormal beats yields:
- Overall classification accuracy: $\text{Acc} = 80.1\%$
- False alarm rate: $\text{FAR} = 0.43\%$
- Average probability of missing a single abnormal beat: $p = 0.199$
- Probability of detecting at least one abnormal beat within the first three consecutive abnormal occurrences: $> 99.2\%$ (computed as $1 - 0.199^3 \approx 1 - 0.0079 = 0.9921$ ).
Knowl 7 — Real-Time Decentralized Structural Health Monitoring Using 1D CNNs
empirical result
In vibration-based Structural Health Monitoring (SHM), compact 1D CNNs enable decentralized damage detection and localization in large civil engineering structures directly from raw ambient acceleration time-series without modal parameter estimation or manual feature extraction.

On a large-scale steel grandstand simulator ( $5\text{ m} \times 6\text{ m}$ footprint with 30 joints):
- Structural damage was introduced by loosening joint bolts, inducing minor changes in rotational stiffness.
- An individual compact 1D CNN was assigned to each sensor location, processing only the local uniaxial or triaxial acceleration signals.
- In both wired setups and triaxial wireless sensor networks (WSN) using shallow architectures (2 CNN layers with 4 neurons each followed by 2 MLP layers with 5 neurons each), the system detected and localized all damaged joints across single- and double-damage scenarios with $100\%$ accuracy (zero false alarms and zero misses).
- On standard multi-core CPU hardware, inference executed at a rate 45 times faster than real-time signal acquisition speed.
Knowl 8 — Motor Current and Bearing Fault Condition Monitoring via Compact 1D CNNs
empirical result
Compact 1D CNNs diagnose bearing faults and motor anomalies by operating directly on raw 1D stator current waveforms or raw vibration signals, performing layered sub-band frequency decomposition through their hidden convolutional kernels to capture subtle mechanical defect signatures.

When compared against conventional feature-extraction pipelines—which combine Wavelet Packet Decomposition (WP) or Fast Fourier Transform (FFT) features with Multi-Layer Perceptrons (MLP), Radial Basis Function Networks (RBFN), or Support Vector Machines (SVM)—the 1D CNN method achieves:
- Near $100\%$ fault detection and classification accuracy, outperforming all conventional feature/classifier combinations on receiver operating characteristic (ROC) curves.
- Substantial reduction in computational latency: total execution time is under $50\text{ ms}$ , whereas conventional methods require over $400\text{ ms}$ (with the vast majority of time consumed by explicit handcrafted feature extraction).
Knowl 9 — Modular Multilevel Converter (MMC) Real-Time Open-Circuit Fault Detection
empirical result
In Modular Multilevel Converters (MMC), compact 1D CNNs perform real-time open-circuit switch fault detection and submodule (SM) fault identification by directly monitoring raw 1D signals of cell capacitor voltages and differential currents.

The compact 1D CNN detection system achieves:
- Fault detection and switch localization accuracy of practically $100\%$ .
- Zero false alarms across varying MMC operational parameters, load currents, and fault timing conditions.
- Detection latency of less than $0.1\text{ s}$ from fault occurrence.
- Scalability to massive multi-cell configurations: large MMC systems containing hundreds of submodules can be partitioned into modular groups (e.g., 8 SMs per group), where identically structured, dedicated 1D CNNs monitor each group in parallel to instantly isolate failing switches.
Knowl 10 — Linear Neuron Homogeneity Limitation and Operational Neural Networks
limitation
A fundamental architectural limitation of conventional 1D and 2D CNNs is their structural and operational homogeneity: every neuron in the network executes the same first-order linear model (linear weighted sums of inputs followed by a fixed scalar activation function) derived from 1950s perceptron formulations. This linear aggregation is an oversimplification of biological neural processing and restricts the capacity of shallow networks to learn highly complex, multi-modal signal patterns without significantly increasing network depth and training data volume.

To overcome this limitation, Generalized Operational Perceptrons (GOPs) and Operational Neural Networks (ONNs) introduce heterogeneous neuron architectures where each neuron can encapsulate non-linear nodal operators (transforming individual synaptic inputs non-linearly) and non-linear pool operators (aggregating transformed inputs). 1D ONNs provide enhanced non-linear representation capabilities, enabling compact networks with few layers to learn complex 1D signal spaces from limited training datasets where conventional 1D CNNs face performance constraints.

Coverage note — Deliberately omitted historical background on 2D CNNs (e.g., LeNet, AlexNet, GoogLeNet) and general literature surveys of bearing fault conversion methods that do not constitute the paper's original formulations or specific experimental contributions.

References

1.A. Cichoki, R. Unbehauen, Neural Networks for Optimization and Signal Processing, 3rd ed., 1994.
2.S.O. Haykin, Neural Networks and Learning Machines, 2008. doi:978-0131471399.
3.S.M. Warren, P. Walter, A logical calculus of the ideas immanent in nervous activity, Bull. Math. Biophys. 5 (1943) 115–133. http://www.cse.chalmers.se/~coquand/AUTOMATA/mcp.pdf.
4.R. F., The perceptron: a probabilistic model for information storage and organization in the brain, Psychol. Rev. 65 (1958) 386.
5.J.P. Resop, A Comparison of Artificial Neural Networks and Statistical Regression with Biological Resources Applications, 2006.
6.T. Ince, S. Kiranyaz, J. Pulkkinen, M. Gabbouj, Evaluation of global and local training techniques over feed-forward neural network architecture spaces for computer-aided medical diagnosis, Expert Syst. Appl. 37 (2010) 8450–8461. doi:10.1016/j.eswa.2010.05.033.
7.S. Kiranyaz, T. Ince, A. Iosifidis, Moncef Gabbouj, Progressive Operational Perceptrons, Neurocomputing. (2016). doi:http://dx.doi.org/10.1016/j.neucom.2016.10.044.
8.S. Kiranyaz, T. Ince, A. Iosifidis, M. Gabbouj, Generalized model of biological neural networks: Progressive operational perceptrons, Proc. Int. Jt. Conf. Neural Networks. 2017–May (2017) 2477–2485. doi:10.1109/IJCNN.2017.7966157.
9.T.W. Rauber, K. Berns, Kernel multilayer perceptron, Proc. - 24th SIBGRAPI Conf. Graph. Patterns Images. (2011) 337–343. doi:10.1109/SIBGRAPI.2011.21.
10.H. Ogai, B. Bhattacharya, Pipe Inspection Robots for Structural Health and Condition Monitoring, 2018. doi:10.1007/978-81-322-3751-8.
11.M.Y. Mashor, Hybrid multilayered perceptron networks, Int. J. Syst. Sci. 31 (2000) 771–785. doi:10.1080/00207720050030815.
12.S. Kiranyaz, T. Ince, A. Yildirim, M. Gabbouj, Evolutionary artificial neural networks by multi-dimensional particle swarm optimization, Neural Networks. 22 (2009) 1448–1462. doi:10.1016/j.neunet.2009.05.013.
13.T. Ince, S. Kiranyaz, M. Gabbou, A generic and robust system for automated patient-specific classification of ECG signals, IEEE Trans. Biomed. Eng. 56 (2009) 1415–1426. doi:10.1109/TBME.2009.2013934.
14.S. Kiranyaz, T. Ince, M. Gabbouj, Multidimensional Particle Swarm Optimization for Machine Learning and Pattern Recognition, Springer, 2014. http://www.springer.com/computer/ai/book/978-3-642-37845-4%5Cnhttps://www.springer.com/computer/ai/book/978-3-642-37845-4.
15.S.A. Mohseni, A.H. Tan, Optimization of neural networks using variable structure systems, IEEE Trans. Syst. Man, Cybern. Part B Cybern. 42 (2012) 1645–1653. doi:10.1109/TSMCB.2012.2197610.
16.D.T. Tran, S. Kiranyaz, M. Gabbouj, A. Iosifidis, Heterogeneous Multilayer Generalized Operational Perceptron, (2018). http://arxiv.org/abs/1804.05093.
17.D.T. Tran, S. Kiranyaz, M. Gabbouj, A. Iosifidis, Progressive Operational Perceptron with Memory, (2018). http://arxiv.org/abs/1808.06377.
18.D.H. Hubel, T.N. Wiesel, Receptive fields of single neurones in the cat’s striate cortex, J. Physiol. 148 (1959) 574–591. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1363130/.
19.D.H. Hubel, Single unit activity in lateral geniculate body and optic tract of unrestrained cats, J. Physiol. 150 (1960) 91–104. doi:10.1113/jphysiol.1960.sp006375.
20.D.H.Hubel, T.N.Wiesel, Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex., J. Physiol. 160 (1962) 106–54. http://www.ncbi.nlm.nih.gov/pubmed/14449617%0Ahttp://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC1359523.
21.D.H.Hubel, T.N.Wiesel, Receptive fields of cells in striate cortex of very young, visually inexperienced kittens, J. Neurophysiol. 26 (1963) 994–1002.
22.D.H. Hubel, T.N. Wiesel, Receptive fields and functional architecture of monkey striate cortex., J. Physiol. 195 (1968) 215–43. http://www.ncbi.nlm.nih.gov/pubmed/4966457%0Ahttp://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC1557912.
23.K. Fukushima, S. Miyake, Neocognitron: A new algorithm for pattern recognition tolerant of deformations and shifts in position, Pattern Recognit. 15 (1982) 455–469. doi:10.1016/0031-3203(82)90024-3.
24.D.E. Rumelhart, G. Hinton, R. J. Williams, Learning representations by back-propagation errors, Nature. 323 (1986) 533–536.
25.Y. LeCun;, B. Boser, J.S. Denker, R.E. Howard, W. Habbard, L.D. Jackel, Handwritten Digit Recognition with a Back-Propagation Network, Adv. Neural Inf. Process. Syst. (1990) 396–404. doi:10.1111/dsu.12130.
26.Y. LeCun, Gradient Based Learning Applied To Document Recognition, Proc. IEEE. 86 (1998) 2278–2324.
27.L. Yann, C. Corinna, B. Chris, MNIST handwritten digit database, New York Univ. (2018). http://yann.lecun.com/exdb/mnist/.
28.A. Krizhevsky, I. Sutskever, G.E. Hinton, ImageNet Classification with Deep Convolutional Neural Networks, ImageNet Classif. with Deep Convolutional Neural Networks. (2012) 1097–1105. http://papers.nips.cc/paper/4824-imagenet-classification-w%5Cnpapers3://publication/uuid/1ECF396A-CEDA-45CD-9A9F-03344449DA2A.
29.J. Deng, W. Dong, R. Socher, L.-J. Li, Kai Li, Li Fei-Fei, ImageNet: A large-scale hierarchical image database, 2009 IEEE Conf. Comput. Vis. Pattern Recognit. (2009) 248–255. doi:10.1109/CVPR.2009.5206848.
30.L. Juan, A Comparison of SIFT , PCA-SIFT and SURF, Int. J. Image Process. 3 (2010) 143–152. http://www.cscjournals.org/csc/manuscript/Journals/IJIP/volume3/Issue4/IJIP-51.pdf.
31.T. Ojala, M. Pietikaeinen, Multiresolution Gray-Scale and Rotation Invariant Texture Classification with Local Binary Patterns, IEEE Trans. Pattern Anal. Mach. Intell. 24 (2005) 971–987.
32.N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov, Dropout: a simple way to prevent neural networks from overfitting, J. Mach. Learn. Res. 1 (2004) 11. doi:10.1016/j.micromeso.2003.09.025.
33.M.D. Zeiler, R. Fergus, Visualizing and understanding convolutional networks, Lect. Notes Comput. Sci. (Including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics). 8689 LNCS (2014) 818–833. doi:10.1007/978-3-319-10590-1_53.
34.O. Abdel-Hamid, A.R. Mohamed, H. Jiang, G. Penn, Applying convolutional neural networks concepts to hybrid NN-HMM model for speech recognition, in: ICASSP, IEEE Int. Conf. Acoust. Speech Signal Process. - Proc., 2012: pp. 4277–4280. doi:10.1109/ICASSP.2012.6288864.
35.Y. Kim, Convolutional Neural Networks for Sentence Classification, in: Proc. 2014 Conf. Empir. Methods Nat. Lang. Process., 2014: pp. 1746–1751. http://arxiv.org/abs/1408.5882.
36.C. Lu, Z. Wang, B. Zhou, Intelligent fault diagnosis of rolling bearing using hierarchical convolutional network based health state classification, Adv. Eng. Informatics. 32 (2017) 139–151. doi:10.1016/j.aei.2017.02.005.
37.X. Ding, Q. He, Energy-Fluctuated Multiscale Feature Learning with Deep ConvNet for Intelligent Spindle Bearing Fault Diagnosis, IEEE Trans. Instrum. Meas. 66 (2017) 1926–1935. doi:10.1109/TIM.2017.2674738.
38.X. Guo, L. Chen, C. Shen, Hierarchical adaptive deep convolution neural network and its application to bearing fault diagnosis, Meas. J. Int. Meas. Confed. 93 (2016) 490–502. doi:10.1016/j.measurement.2016.07.054.
39.O. Janssens, V. Slavkovikj, B. Vervisch, K. Stockman, M. Loccufier, S. Verstockt, R. Van de Walle, S. Van Hoecke, Convolutional Neural Network Based Fault Detection for Rotating Machinery, J. Sound Vib. 377 (2016) 331–345. doi:10.1016/j.jsv.2016.05.027.
40.D.-T. Hoang, H.-J. Kang, Convolutional Neural Network Based Bearing Fault Diagnosis, in: D.-S. Huang, K.-H. Jo, J.C. Figueroa-García (Eds.), Intell. Comput. Theor. Appl., Springer International Publishing, Cham, 2017: pp. 105–111.
41.Z. Wei, P. Gaoliang, L. Chuanhao, Bearings Fault Diagnosis Based on Convolutional Neural Networks with 2- D Representation of Vibration Signals as Input, 13001 (2017) 1–5.
42.D.K. Appana, W. Ahmad, J.-M. Kim, Speed Invariant Bearing Fault Characterization Using Convolutional Neural Networks, (2017) 189–198. doi:10.1007/978-3-319-69456-6_16.
43.S. Li, G. Liu, X. Tang, J. Lu, J. Hu, An Ensemble Deep Convolutional Neural Network Model with Improved D-S Evidence Fusion for Bearing Fault Diagnosis, Sensors. 17 (2017) 1729. doi:10.3390/s17081729.
44.K.B. Lee, S. Cheon, C.O. Kim, L. Dalfino, F. Puntillo, A. Mosca, R. Monno, M. Luigia, S. Coppolecchia, G. Miragliotta, F. Bruno, D. Verstraete, M. Engineering, M. Engineering, M.F. Guo, X.D. Zeng, D.Y. Chen, N.C. Yang, An adaptive deep convolutional neural network for rolling bearing fault diagnosis, Hindawi Shock Vib. 30 (2017) 1–29. doi:10.1109/TSM.2017.2676245.
45.J. Ruiz, J. Pérez, J. Blázquez, Arrhythmia Detection Using Convolutional Neural Models, in: Int. Symp. Distrib. Comput. Artif. Intell., 2018.
46.M. Zihlmann, D. Perekrestenko, M. Tschannen, Convolutional recurrent neural networks for electrocardiogram classification, Computing. (2017).
47.S. Kiranyaz, T. Ince, R. Hamila, M. Gabbouj, Convolutional Neural Networks for patient-specific ECG classification, in: Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. EMBS, 2015. doi:10.1109/EMBC.2015.7318926.
48.S. Kiranyaz, T. Ince, M. Gabbouj, Real-Time Patient-Specific ECG Classification by 1-D Convolutional Neural Networks, IEEE Trans. Biomed. Eng. 63 (2016) 664–675. doi:10.1109/TBME.2015.2468589.
49.S. Kiranyaz, T. Ince, M. Gabbouj, Personalized Monitoring and Advance Warning System for Cardiac Arrhythmias, Sci. Rep. 7 (2017). doi:10.1038/s41598-017-09544-z.
50.O. Avci, O. Abdeljaber, S. Kiranyaz, M. Hussein, D.J. Inman, Wireless and real-time structural damage detection: A novel decentralized method for wireless sensor networks, (2018). doi:10.1016/j.jsv.2018.03.008.
51.O. Avci, O. Abdeljaber, S. Kiranyaz, D. Inman, Structural Damage Detection in Real Time: Implementation of 1D Convolutional Neural Networks for SHM Applications, in: C. Niezrecki (Ed.), Struct. Heal. Monit. Damage Detect. Vol. 7 Proc. 35th IMAC, A Conf. Expo. Struct. Dyn. 2017, Springer International Publishing, Cham, 2017: pp. 49–54. doi:10.1007/978-3-319-54109-9_6.
52.O. Abdeljaber, O. Avci, S. Kiranyaz, M. Gabbouj, D.J. Inman, Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks, J. Sound Vib. 388 (2017). doi:10.1016/j.jsv.2016.10.043.
53.O. Avci, O. Abdeljaber, S. Kiranyaz, B. Boashash, H. Sodano, D.J. Inman, Efficiency Validation of One Dimensional Convolutional Neural Networks for Structural Damage Detection Using a SHM Benchmark Data, 25th Int. Congr. Sound Vib. (2018).
54.O. Abdeljaber, O. Avci, M.S. Kiranyaz, B. Boashash, H. Sodano, D.J. Inman, 1-D CNNs for structural damage detection: Verification on a structural health monitoring benchmark data, Neurocomputing. (2017). doi:10.1016/j.neucom.2017.09.069.
55.T. Ince, S. Kiranyaz, L. Eren, M. Askar, M. Gabbouj, Real-Time Motor Fault Detection by 1-D Convolutional Neural Networks, IEEE Trans. Ind. Electron. 63 (2016) 7067–7075. doi:10.1109/TIE.2016.2582729.
56.S. Kiranyaz, A. Gastli, L. Ben-Brahim, N. Alemadi, M. Gabbouj, Real-Time Fault Detection and Identification for MMC using 1D Convolutional Neural Networks, IEEE Trans. Ind. Electron. (2018). doi:10.1109/TIE.2018.2833045.
57.O. Abdeljaber, S. Sassi, O. Avci, S. Kiranyaz, A.A. Ibrahim, M. Gabbouj, Fault Detection and Severity Identification of Ball Bearings by Online Condition Monitoring, IEEE Trans. Ind. Electron. (2018) 1–1. doi:10.1109/tie.2018.2886789.
58.L. Eren, T. Ince, S. Kiranyaz, A Generic Intelligent Bearing Fault Diagnosis System Using Compact Adaptive 1D CNN Classifier, J. Signal Process. Syst. 91 (2019) 179–189. doi:10.1007/s11265-018-1378-3.
59.L. Eren, Bearing fault detection by one-dimensional convolutional neural networks, Math. Probl. Eng. 2017 (2017). doi:10.1155/2017/8617315.
60.W. Zhang, C. Li, G. Peng, Y. Chen, Z. Zhang, A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load, Mech. Syst. Signal Process. 100 (2018) 439–453. doi:10.1016/j.ymssp.2017.06.022.
61.H.G. E., S.R. R., Reducing the dimensionality of data with neural networks, Science (80-. ). 313 (2006) 504–507.
62.U.R. Acharya, H. Fujita, O.S. Lih, Y. Hagiwara, J.H. Tan, M. Adam, Automated detection of arrhythmias using different intervals of tachycardia ECG segments with convolutional neural network, Inf. Sci. (Ny). (2017). doi:10.1016/j.ins.2017.04.012.
63.Z. Xiong, M. Stiles, J. Zhao, Robust ECG Signal Classification for Detection of Atrial Fibrillation Using a Novel Neural Network, in: 2017 Comput. Cardiol., 2017.
64.U.R. Acharya, S.L. Oh, Y. Hagiwara, J.H. Tan, M. Adam, A. Gertych, R.S. Tan, A deep convolutional neural network model to classify heartbeats, Comput. Biol. Med. (2017). doi:10.1016/j.compbiomed.2017.08.022.
65.Structural Dynamics Application Website, http://structuralvibration.com/.
66.Y. Yu, C. Wang, X. Gu, J. Li, A novel deep learning-based method for damage identification of smart building structures, Struct. Heal. Monit. 18 (2019) 143–163. doi:https://doi.org/10.1177/1475921718804132.
67.H. Khodabandehlou, G. Pekcan, M.S. Fadali, Vibration-based structural condition assessment using convolution neural networks, Struct. Control Heal. Monit. (n.d.). doi:10.1002/stc.2308.
68.Y.J. Cha, W. Choi, O. Büyüköztürk, Deep Learning-Based Crack Damage Detection Using Convolutional Neural Networks, Comput. Civ. Infrastruct. Eng. 32 (2017) 361–378. doi:10.1111/mice.12263.
69.N.S. Gulgec, M. Takáč, S.N. Pakzad, Structural damage detection using convolutional neural networks, Conf. Proc. Soc. Exp. Mech. Ser. 3 Part F2 (2017) 331–337. doi:10.1007/978-3-319-54858-6_33.
70.O. Abdeljaber, A. Younis, O. Avci, N. Catbas, M. Gul, O. Celik, H. Zhang, Dynamic Testing of a Laboratory Stadium Structure, in: Geotech. Struct. Eng. Congr. 2016, n.d.: pp. 1719–1728. doi:10.1061/9780784479742.147.
71.O. Avci, O. Abdeljaber, S. Kiranyaz, D.J. Inman, Convolutional Neural Networks for Real-time and Wireless Damage Detection, in: IMAC XXXVII, Int. Modal Anal. Conf., Springer International Publishing, Orlando, FL, USA, 2019.
72.S. Dyke, D. Bernal, J. Beck, C. Ventura, Experimental phase II of the structural health monitoring benchmark problem, in: Proc. 16th ASCE Eng. Mech. Conf., 2003: pp. 1–7.
73.S. Sassi, B. Badri, M. Thomas, Tracking surface degradation of ball bearings by means of new time domain scalar indicators, Int. J. COMADEM. 11 (2008) 36–45.
74.A. Mohamed, S. Sassi, M.R. Paurobally, Model-based analysis of gears’ dynamic behavior in the presence of multiple cracks, J. Shock Vib. (2018).
75.M. Blodt, P. Granjon, B. Raison, G. Rostaing, Models for bearing damage detection in induction motors using stator current monitoring, IEEE Trans. Ind. Electron. 55 (2008) 1813–1822. doi:10.1109/TIE.2008.917108.
76.L. Eren, M.J. Devaney, Bearing Damage Detection via Wavelet Packet Decomposition of the Stator Current, IEEE Trans. Instrum. Meas. 53 (2004) 431–436. doi:10.1109/TIM.2004.823323.
77.L. Eren, A. Karahoca, M.J. Devaney, Neural network based motor bearing fault detection, Conf. Rec. - IEEE Instrum. Meas. Technol. Conf. 3 (2004) 1657–1660. doi:10.1109/IMTC.2004.1351399.
78.G.F. Bin, J.J. Gao, X.J. Li, B.S. Dhillon, Early fault diagnosis of rotating machinery based on wavelet packets - Empirical mode decomposition feature extraction and neural network, Mech. Syst. Signal Process. 27 (2012) 696–711. doi:10.1016/j.ymssp.2011.08.002.
79.G. Serpen, Z. Gao, Complexity analysis of multilayer perceptron neural network embedded into a wireless sensor network, Procedia Comput. Sci. 36 (2014) 192–197. doi:10.1016/j.procs.2014.09.078.
80.S. Fan, Giving Neural Nets an Innate Brain-Like Structure Could Bolster Deep Learning, (n.d.). https://into.ai/blog/news-stories/giving-neural-nets-an-innate-brain-like-structure-could-bolster-deep-learning-4/.
81.S. Kiranyaz, T. Ince, A. Iosifidis, M. Gabbouj, Operational Neural Networks, (2019). http://arxiv.org/abs/1902.11106.

Citation

MLA

Kiranyaz, S., et al. “1D Convolutional Neural Networks and Applications: A Survey”. Mechanical Systems and Signal Processing, vol. 151, 2021, p. 107398, https://doi.org/10.1016/j.ymssp.2020.107398.

APA

Kiranyaz, S., Avci, O., Abdeljaber, O., Ince, T., Gabbouj, M., & Inman, D. J. (2021). 1D convolutional neural networks and applications: A survey. Mechanical Systems and Signal Processing, 151, 107398. https://doi.org/10.1016/j.ymssp.2020.107398

Chicago

Kiranyaz, S., O. Avci, O. Abdeljaber, T. Ince, M. Gabbouj, and D. J. Inman. 2021. “1D Convolutional Neural Networks and Applications: A Survey”. Mechanical Systems and Signal Processing 151: 107398. https://doi.org/10.1016/j.ymssp.2020.107398.

Harvard

Kiranyaz, S. et al. (2021) “1D convolutional neural networks and applications: A survey”, Mechanical Systems and Signal Processing, 151, p. 107398. Available at: https://doi.org/10.1016/j.ymssp.2020.107398.

Vancouver

1. Kiranyaz S, Avci O, Abdeljaber O, Ince T, Gabbouj M, Inman DJ (2021) 1D convolutional neural networks and applications: A survey. Mechanical Systems and Signal Processing 151:107398

BibTeX

@article{Kiranyaz_2021, title={1D convolutional neural networks and applications: A survey}, volume={151}, ISSN={0888-3270}, url={http://dx.doi.org/10.1016/j.ymssp.2020.107398}, DOI={10.1016/j.ymssp.2020.107398}, journal={Mechanical Systems and Signal Processing}, publisher={Elsevier BV}, author={Kiranyaz, Serkan and Avci, Onur and Abdeljaber, Osama and Ince, Turker and Gabbouj, Moncef and Inman, Daniel J.}, year={2021}, month=Apr, pages={107398} }

Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

1D Convolutional Neural Networks and Applications: A Survey

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Compact 1D Convolutional Neural Network Architecture for Raw Sequential Signals

Knowl 2 — Mathematical Formulation of Forward Propagation in 1D CNN Layers

Knowl 3 — Back-Propagation and Parameter Sensitivities in 1D CNN Layers

Knowl 4 — Iterative Back-Propagation Training Algorithm for 1D CNNs

Knowl 5 — Computational Complexity Analysis of Forward and Backward Propagation in 1D CNNs

Knowl 6 — Patient-Specific and Personalized Early ECG Arrhythmia Detection Using 1D CNNs

Knowl 7 — Real-Time Decentralized Structural Health Monitoring Using 1D CNNs

Knowl 8 — Motor Current and Bearing Fault Condition Monitoring via Compact 1D CNNs

Knowl 9 — Modular Multilevel Converter (MMC) Real-Time Open-Circuit Fault Detection

Knowl 10 — Linear Neuron Homogeneity Limitation and Operational Neural Networks

References

Citation

BibTeX

Access the Paper