Calibration Block Research Articles

Ultrasonic Total Focusing Method Technology in Carbon Steel Butt Weld Application

Carbon steel butt welds are widely used in aircraft carrier, shipbuilding, railway, aerospace and other industries. As a new technology of ultrasonic non-destructive testing, Total Focusing Method (TFM) technology is a method of image reconstruction in which the value of each constituent datum of the image results from focused ultrasound. TFM may also be understood as a broad term encompassing a family of processing techniques for image reconstruction from full matrix capture (FMC). This paper compares resolution and focus between TFM and conventional phased array technology, by using phased array test block type B. The paper also introduces the testing system, including equipment, probe, wedge, calibration block as per ASME, PC analysis software, for detection of 21mm thickness carbon steel welds. In order to improve the defect detection ability, a customized software on board was designed based on TFM technology, and an appropriate propagation mode was adopted for different types of defects, such as longitudinal cracks on surface, lack of fusion, porosities, slag inclusions. The test results showing that compared with conventional phased array technology, the defect information such as length and depth detected by TFM are nearer to the actual situation, and the shape is closer to the physical one. TFM technology has more advantages in quantitative and qualitative capabilities in testing carbon steel butt welds, and can meet the requirements of on-site testing.

3D reconstruction and volume measurement of irregular objects based on RGB-D camera

Abstract To address the challenge of measuring volumes of irregular objects, this paper proposes a volume measurement method based on 3D point cloud reconstruction. The point clouds of the object with multiple angles are obtained from an RGB-D camera mounted on a robotic arm, and then are reconstructed to form a whole complete point cloud to calculate the volume of the object. Firstly, the robotic arm is controlled to move to four angles for capturing the original point clouds of the target. Then, by using the rotation and translation matrices obtained from the calibration block pre-registration, the point clouds data from the four angles are fused and reconstructed. Subsequently, the issue of missing bottom point cloud data is addressed using a bottom-filling algorithm. Following this, the efficiency of the point cloud volume calculation algorithm is enhanced through the application of AABB filtering. Finally, the reconstructed point cloud volume is calculated using a slicing algorithm that integrates 2D point cloud segmentation and point cloud sorting. Experimental results show that this method achieves a volume measurement accuracy of over 95% for irregular objects and exhibits good robustness.

A Multi-Branch Feature Extraction Residual Network for Lightweight Image Super-Resolution

Single-image super-resolution (SISR) seeks to elucidate the mapping relationships between low-resolution and high-resolution images. However, high-performance network models often entail a significant number of parameters and computations, presenting limitations in practical applications. Therefore, prioritizing a light weight and efficiency becomes crucial when applying image super-resolution (SR) to real-world scenarios. We propose a straightforward and efficient method, the Multi-Branch Feature Extraction Residual Network (MFERN), to tackle lightweight image SR through the fusion of multi-information self-calibration and multi-attention information. Specifically, we have devised a Multi-Branch Residual Feature Fusion Module (MRFFM) that leverages a multi-branch residual structure to succinctly and effectively fuse multiple pieces of information. Within the MRFFM, we have designed the Multi-Scale Attention Feature Fusion Block (MAFFB) to adeptly extract features via convolution and self-calibration attention operations. Furthermore, we introduce a Dual Feature Calibration Block (DFCB) to dynamically fuse feature information using dynamic weight values derived from the upper and lower branches. Additionally, to overcome the limitation of convolution in solely extracting local information, we incorporate a Transformer module to effectively integrate global information. The experimental results demonstrate that our MFERN exhibits outstanding performance in terms of model parameters and overall performance.

YOLOv7-Based Intelligent Weed Detection and Laser Weeding System Research: Targeting Veronica didyma in Winter Rapeseed Fields

In recent years, rapeseed oil has received considerable attention in the agricultural sector, experiencing appreciable growth. However, weed-related challenges are hindering the expansion of rapeseed production. This paper outlines the development of an intelligent weed detection and laser weeding system—a non-chemical and precision agricultural protection method of weeding Veronica didyma in winter rapeseed fields in the Yangtze River Basin. A total of 234 Veronica didyma images were obtained to compile a database for a deep-learning model, and YOLOv7 was used as the detection model for training. The effectiveness of the model was demonstrated, with a final accuracy of 94.94%, a recall of 95.65%, and a mAP@0.5 of 0.972 obtained. Subsequently, parallel-axis binocular cameras were selected as the image acquisition platform, with binocular calibration and semi-global block matching used to locate Veronica didyma within a cultivation box, yielding a minimum confidence and camera height values of 70% and 30 cm, respectively. The intelligent weed detection and laser weeding system was then built, and the experimental results indicated that laser weeding was practicable with a 100 W power and an 80 mm/s scanning speed, resulting in visibly lost activity in Veronica didyma and no resprouting within 15 days of weeding. The successful execution of Veronica didyma detection and laser weeding provides a new reference for the precision agricultural protection of rapeseed in winter and holds promise for its practical application in agricultural settings.

Advanced Excitation Symmetry Analysis Method (ESAM) for improved automated ultrasonic testing of polished materials

One of the strategic plans of the international NDT community is to define standards for developing advanced and digitalized non-destructive testing for automated set-up in mass production [1,2]. Excitation Symmetry Analysis Method (ESAM) are new coded signal processing tools exploring symmetry properties [3]. ESAM method aims to extract the nonlinearity from an ultrasound output by studying the property of symmetry of the transfer function. More precisely, the extraction of nonlinearity is done using basis function associated to a group which is associated to an a priori transfer function. As soon as the linear tomography uses the invariant properties of amplitude dependence, nonlinear tomography should exploit the invariance properties (or symmetry properties) of the complex system. Consequently, the invariance of the stationary properties of a complex medium would be supposed to be associated to a signature of the degradation, that will be extracted with advanced signal processing tools. The idea remains the same: studying the symmetry and more precisely the time reversal one and the inversion. A signal processing is used to measure discrepancies in the autocorrelation function (output signal called chirp coda) for two inputs: a signal and its inversion (pulse inversion). The experiments were done on different sample with and without cracks. It has been shown that when the nonlinearities are extracted from the chirp coda amplitude are non- negligible in crack area and nonlinearities are out of phase. Then, mixing symmetries like time reversal and pulse inversion enable the detection of defect that are difficult to detect with conventional ultrasound method. Indeed, for such crack a destructive mechanical analysis called Jomini is done where the material is polished and examined with microscope. The sensitivity improvement of chirp-coded signal processing has been validated in various domains. Coded excitation techniques, used in communication systems such as radar and sonar provide improved SNR without increasing the amplitude of excitation. The typical test equipment consists of a preamplifier Juvitek TRA-02 (0.02 - 5 MHz) connected to a computer, an amplifier ENI model A150 (55 dB at 0.3-35 MHz), a shear wave transducer Technisonic (2.25 MHz), and a longitudinal wave transducer Panametrics V155 (5 MHz). The experimental device was tested with the V3 calibration block, improved, and specially scaled in order to access to a wide range of multivalued parameters: mechanical properties, ultrasonic parameters (celerity and attenuation) and local geometric data.

Synthetic phantoms enable calibration between abattoir based dual energy X-ray absorptiometers used for prediction of lamb carcass composition

Dual energy x-ray absorptiometry (DXA) devices were installed at two Australian abattoirs to predict computed tomography (CT) determined fat % and lean % of lamb carcasses. This study tested three algorithms developed for these devices, termed β1, β2 and β3, and assessed their accuracy and precision in predicting CT composition. Algorithm β3 included the use of a plastic phantom calibration block scanned by both DXA devices to adjust prediction equations, resulting in superior accuracy to the algorithms without phantom calibration (β1 and β2). When compared to the gold-standard CT composition, the bias of the DXA predictions was lowest when using algorithm β3 at the two sites (−1.17%, −0.49% for fat %, 0.11%, −0.37% for lean %). The difference of DXA composition predictions between sites was lowest when using algorithm β3, which demonstrated between site differences of 0.59 CT fat %, and 0.46 CT lean%. In contrast, algorithm β1 and β2 produced differences of 23.7% and 30.8% for CT fat, and 17.3% and 21.9% for CT lean between the two DXA devices. There was a small difference of 0.78% between the fat predictions of the first DXA image compared to the second DXA image for each carcass. The precision of predictions improved slightly using algorithm β3. This work demonstrates that the in-line DXA systems can produce comparable results across sites.

CIVA Modelling Module for Zonal Discrimination Method Part 2-System Qualification Aspects

Automatic Ultrasonic Testing (AUT) systems for pipeline girth weld inspections are usually subjected to a qualification process. This process consists of assessment of performance capabilities including detection, sizing, temperature range and repeatability. One of the most demanding qualification programmes is described in DNV ST-F101 Appendix E [1]. With the introduction of the AUT module in the 2023 edition of CIVA simulation software, the incorporation of metamodeling allows users to predict detection reliability via probability of detection (POD) and sizing accuracy of the zonal discrimination method (ZDM) using amplitude apportioning. Part 1 of this project demonstrated the ability of CIVA to simulate the calibration block and strip chart output. In Part 2 we illustrate the techniques that can be used to predict POD and sizing accuracy as might be done in a system qualification. Results compare well to field data.

Guidance on configuring volumetric targets for AUT using CIVA

The main standards relating to AUT (Automated Ultrasonic Testing) using zonal discrimination have long had requirements to incorporate separate channels dedicated to detecting volumetric flaws. However, none of the standards specify how the beams are to be configured. The instructions are quite generic indicating that the channels are to ensure the complete volumetric examination of the weld through-thickness. Scattering flat bottom hole targets in a calibration block such that they evenly distribute in the volume is not a guarantee that the target placement provides the required complete volume coverage. This paper illustrates how the Coverage component of the CIVA AUT module helps to identify the best placement of volumetric targets in an AUT Zonal Discrimination procedure and prevents target redundancy.

A Brief Review of the Literature for Published Dual-Energy X-Ray Absorptiometry Protocols for Athletes.

Dual-energy x-ray absorptiometry (DXA) is widely known for its utility in diagnosing a patient with osteopenia or osteoporosis; however, its utility in evaluation of body composition and potential athletic performance was previously routinely overlooked. In recent years, athletic programs have begun employing this equipment during athlete screening. However, it is currently unknown how athletic programs are utilizing this information to guide an athlete's training and health. To explore the literature to identify what is known regarding sports performance and athlete health based on body composition recordings. PubMed and Google Scholar databases were searched for this brief review. A focus was placed on articles within the past 10 years that discussed DXA protocols within athletic populations; 14 articles were included in this brief literature review. Brief literature review. Level 5. One member of the research team searched the literature and retrieved articles with the purpose of analyzing and/or explaining DXA imaging in body composition analysis of active persons (primarily athletes). Quality assurance scans with a phantom calibration block as well as athlete prescreening condition and activity standardization was routinely recommended. However, only 1 study reported a specific DXA protocol for athletes, and only 1 study described guidelines for how to report DXA results in athletic populations, suggesting it is plausible yet difficult due to the small changes detectable. Due to the limited literature as well as a lack of reference values for specific athletic populations, the authors of this review recommend using the current Nana et al (Int J Sport Nutr Exerc Metab 2015;25:198-215) DXA protocol for performing DXA scans in the athletic population as well as current Hind et al (J Clin Densitom 2018;21:429-443) guidelines for distributing the information.

CKD-TransBTS: Clinical Knowledge-Driven Hybrid Transformer with Modality-Correlated Cross-Attention for Brain Tumor Segmentation.

Brain tumor segmentation (BTS) in magnetic resonance image (MRI) is crucial for brain tumor diagnosis, cancer management and research purposes. With the great success of the ten-year BraTS challenges as well as the advances of CNN and Transformer algorithms, a lot of outstanding BTS models have been proposed to tackle the difficulties of BTS in different technical aspects. However, existing studies hardly consider how to fuse the multi-modality images in a reasonable manner. In this paper, we leverage the clinical knowledge of how radiologists diagnose brain tumors from multiple MRI modalities and propose a clinical knowledge-driven brain tumor segmentation model, called CKD-TransBTS. Instead of directly concatenating all the modalities, we re-organize the input modalities by separating them into two groups according to the imaging principle of MRI. A dual-branch hybrid encoder with the proposed modality-correlated cross-attention block (MCCA) is designed to extract the multi-modality image features. The proposed model inherits the strengths from both Transformer and CNN with the local feature representation ability for precise lesion boundaries and long-range feature extraction for 3D volumetric images. To bridge the gap between Transformer and CNN features, we propose a Trans&CNN Feature Calibration block (TCFC) in the decoder. We compare the proposed model with six CNN-based models and six transformer-based models on the BraTS 2021 challenge dataset. Extensive experiments demonstrate that the proposed model achieves state-of-the-art brain tumor segmentation performance compared with all the competitors.

Comparative study of advanced image reconstruction algorithms for complex arbitrary components

Accurate ultrasonic imaging and inspection of specimens with non-planar complex surfaces are often challenging from both transducer coupling and imaging perspectives. Several inspection strategies have been developed to inspect such complex interfaced components such as flat arrays with flexible wedges or water-filled chambers with a flexible membrane. In these methods, firstly surface profiles are estimated and then Fermat’s principle or other algorithms are applied to calculate delay laws. The main drawback of these methods is the longer computation time and therefore real time inspection is limited, also a coupling problem due to the complex geometries. Alternative approaches have been developed to compute the ultrasonic beam paths such as the Fast-Marching Method (FMM), Dijkstra’s algorithms, numerical analysis, bisection and root-finding algorithms for the inspection of complex surface components. These algorithms can be applied in two and three-dimensional domains resulting in high computation time. This work presents an experimental comparison of an iterative beamforming method for arbitrary shaped surface profile imaging against both Full Matrix Capture-Total Focusing Method (FMC-TFM) and CIVA simulation on an aluminium calibration block with Side drilled holes (SDHs) at various depths in the range of 15-80 mm. The performance of the iterative beamforming method is evaluated in terms of sensitivity, Signal to Noise Ratio (SNR) and Array Performance Indicators (API). The method provides a 2 dB improved SNR when compared to FMC-TFM.

An error compensation controller for milling robots

This paper presents a method of controlling a serial robot for milling by an inverse kinematic controller combined with an outer PD loop (Inverse Dynamics + PD controller), with calibration and compensation of errors in calculating the cutting forces. Because the cutting forces are generated at the time of cutting, at the contact area between the workpiece and the cutting tool, the generalized forces of the cutting forces in the differential equations of motion of robot is always variable and difficult to determine precisely. The cutting forces depend on the cutting mode, the geometric parameters of the cutting layer, the cutting conditions, etc. This study shows an inverse dynamic controller with the outer PD loop and an additional calibration block to compensate the differences between the actual cutting forces and calculated cutting forces (which are caculated by the empirical formula). The cutting forces at each machining time of the calibration block is determined based on the differential equation of motion. The efficiency (convergence time and accuracy) of the proposed controller is evaluated by comparison between the numerical simulation results of the controller with cutting force calibration and the conventional PD controller. In the conventional PD controller, the dynamic model of the robot is assumed to define precisely. The results contribute to design and manufacture the controllers for robotic milling, and to improve the quality of the machined surface.

MFCNet: A multi-modal fusion and calibration networks for 3D pancreas tumor segmentation on PET-CT images

In clinical diagnosis, positron emission tomography and computed tomography (PET-CT) images containing complementary information are fused. Tumor segmentation based on multi-modal PET-CT images is an important part of clinical diagnosis and treatment. However, the existing current PET-CT tumor segmentation methods mainly focus on positron emission tomography (PET) and computed tomography (CT) feature fusion, which weakens the specificity of the modality. In addition, the information interaction between different modal images is usually completed by simple addition or concatenation operations, but this has the disadvantage of introducing irrelevant information during the multi-modal semantic feature fusion, so effective features cannot be highlighted. To overcome this problem, this paper propose a novel Multi-modal Fusion and Calibration Networks (MFCNet) for tumor segmentation based on three-dimensional PET-CT images. First, a Multi-modal Fusion Down-sampling Block (MFDB) with a residual structure is developed. The proposed MFDB can fuse complementary features of multi-modal images while retaining the unique features of different modal images. Second, a Multi-modal Mutual Calibration Block (MMCB) based on the inception structure is designed. The MMCB can guide the network to focus on a tumor region by combining different branch decoding features using the attention mechanism and extracting multi-scale pathological features using a convolution kernel of different sizes. The proposed MFCNet is verified on both the public dataset (Head and Neck cancer) and the in-house dataset (pancreas cancer). The experimental results indicate that on the public and in-house datasets, the average Dice values of the proposed multi-modal segmentation network are 74.14% and 76.20%, while the average Hausdorff distances are 6.41 and 6.84, respectively. In addition, the experimental results show that the proposed MFCNet outperforms the state-of-the-art methods on the two datasets.

Compensation for attenuation on ultrasonic testing at TF coil structure of ITER

The ITER Toroidal Field Coil (TFC) is a massive 16 m × 9 m “D”-shape-superconducting magnet composed of winding pack (WP) and TF coil structure (TFCS). The TFCS is made of thick austenitic stainless steel with over 100 mm. The ultrasonic testing (UT) is used to the welding examination. The longitudinal wave is selected since it is generally used for UT of austenitic stainless-steel welds. There is a concern about the attenuation of the signal level of UT, because the ultrasonic wave is scattered through the weld metal of austenitic stainless steel. This paper reports the following compensation method against the attenuation of the UT signal with actual data applied for ITER TFCS. In the ITER TFCS fabrication, the attenuation of the UT beam in the weld is compensated by the transfer correction factor obtained from a test piece made of the actual TFCS material and weld metal during calibration. UT is performed using distance amplitude characteristic (DAC) curve obtained with a calibration block made from the base material of TFCS with side drill holes (SDHs). In order to capture the sensitivity difference between base material and weld metal, signal obtained with a reference block made from welding joint of TFCS with the same SDHs in the weld are plotted on the DAC. The sensitivity difference is obtained by difference of signal level on DAC and the plot. Then, relationship between the sensitivity difference and distance propagating in the weld metal is obtained. The sensitivity difference is used for assessment of UT signal level to correct decreasing by weld metal selecting proper sensitivity difference with respect to distance propagating in the weld metal in actual UT. This paper includes test results obtained by manufacturing companies using the actual UT probes for TFCS fabrication. The paper also provides obtained correlation between propagation distance in the weld metal and sensitivity difference.

Phase-locked Time-shift Data Augmentation Method for SSVEP Brain-Computer Interfaces.

Steady-state visual evoked potential (SSVEP) based brain-computer interfaces (BCIs) have achieved an information transfer rate (ITR) of over 300 bits/min, but abundant training data is required. The performance of SSVEP algorithms deteriorates greatly under limited data, and the existing time-shift data augmentation method fails to improve it because the phase-locked requirement between training samples is violated. To address this issue, this study proposes a novel augmentation method, namely phase-locked time-shift (PLTS), for SSVEP-BCI. The similarity between epochs at different time moments was evaluated, and a unique time-shift step was calculated for each class to augment additional data epochs in each trial. The results showed that the PLTS significantly improved the classification performance of SSVEP algorithms on the BETA SSVEP datasets. Moreover, under the condition of one calibration block, by slightly prolonging the calibration duration (from 48 s to 51.5 s), the ITR increased from 40.88 ±4.54 bits/min to 122.61 ± 7.05 bits/min with the PLTS. This study provides a new perspective on augmenting data epochs for training-based SSVEP-BCI, promotes the classification accuracy and ITR under limited training data, and thus facilitates the real-life applications of SSVEP-based brain spellers.

Binocular Structured Light 3-D Reconstruction System for Low-Light Underwater Environments: Design, Modeling, and Laser-Based Calibration

The three dimensional (3D) perception of the low-light underwater environment has always been a major challenge which greatly limits the underwater operations. In this paper, an underwater active vision measurement system based on binocular structured light is designed to achieve high precision 3D reconstruction. Firstly, the fusion technology of binocular camera and laser addresses underwater optical attenuation and feature sparsity. Then, in order to avoid the huge inertia caused by the overall movement of the system, a laser scanner based on the mirror-galvanometer is used to accomplish static scanning of the scene. Subsequently, considering the influence of multiple media, underwater refraction models including monocular imaging model, binocular ranging model, and binocular polar curve constraint model are systematically proposed. What’s more, the conventional checkerboard-based passive visual calibration method is ineffective for low-light waters. Therefore, a simple calibration block is designed, and a new multi-objective laser-based calibration algorithm based on laser geometric constraints is proposed. Finally, the effectiveness of the system is verified by analyzing the 3D reconstruction results of underwater objects.

A Power-Efficient CMOS Multi-Band Phased-Array Receiver Covering 24–71-GHz Utilizing Harmonic-Selection Technique With 36-dB Inter-Band Blocker Tolerance for 5G NR

This article introduces a power-efficient 24.25– 71-GHz multi-band phased-array receiver supporting all allocated fifth-generation mobile network new radio (5G NR) frequency range 2 (FR2) bands at 24/28/39/47 GHz and the potential 5G NR-U bands in unlicensed 57–71 GHz. A novel harmonic-selection technique is introduced to extend the operating bandwidth with low power consumption. By switching between the fundamental-selected mode, the second-harmonic-selected mode, and the third-harmonic-selected mode, only signals in the desired bands can be preserved, while the unselected mixing components are rejected. A dual-mode multi-band low-noise amplifier (LNA) based on a configurable transformer is adopted to realize broadband operation with minimized power consumption and noise figure (NF). The Hartley architecture is employed to further improve the image rejection performance. A hybrid-type polyphase filter (PPF) with a detector-based high-precision calibration block is utilized in this work to realize the Hartley operation with reduced insertion loss (IL). The proposed phased-array receiver is fabricated in a standard 65-nm bulk CMOS process. With the concerted efforts of all components, the proposed multi-band receiver can support 5G standard-compliant OFDMA-mode modulated signals up to 256QAM with a 400-MHz channel bandwidth from 24 to 71 GHz. Better than 36-dB inter-band blocker rejections can be maintained by this work. With existing of 0-dBc inter-band blockers at worst case frequencies, this receiver shows EVMs of −33.3, −30.9, −31.6, and −28.5 dB at 28, 39, 47.2, and 60.1 GHz, respectively. The power consumptions for a single receiver channel are 36, 32, 51, and 71 mW at 28, 39, 47.2, and 60.1 GHz, respectively.

FSRDD: An Efficient Few-Shot Detector for Rare City Road Damage Detection

Road damage detection (RDD) is indispensable for safe autonomous driving. Existing RDD models focus on designing feature representations following expert knowledge. However, collecting and labeling all types of samples is time-consuming and leads to insufficient training data. To alleviate the adverse effect of few training samples, a novel few-shot road damage detector (FSRDD) is proposed in this paper to detect rare road damages. The proposed FSRDD includes three stages. First, fully annotated abundant base classes are leveraged to train a base detector, where ghost attention (GA) and proposal feature metric (PFM) modules are developed to eliminate the redundant information and measure the proposal features, respectively. Second, the recognition branch of the detector is fine-tuned using a few samples of all classes. Finally, the test set is inferred with the help of an offline scale-aware prototypical calibration block (SPCB). Extensive experiments show that our FSRDD achieves 10-shot rare road damage detection with 33.4% and 12.9% mAP50 on RDD and CNRDD datasets, respectively, significantly outperforming state-of-the-art methods.

Application of Benford's Law for Quality Assessment of Preventive Screening Data

An empirical Benford's law which describes the probability of the appearance of certain first significant digits in many distributions taken from real life, is used to identify anomalies in various kinds of data. Our aim was to test Benford's law to assess the quality of mass preventive screening data on the example of bioelectrical impedance analysis (BIA) data from Moscow health centers. As was shown earlier, such a data is characterized by a high level of contamination by artificially generated and falsified data. A generated 2010–2019 database of BIA measurements contained 1361019 measurement records in the age range of the examined persons from 5 to 96 years. Application of the expert quality assessment algorithm, which was used as a reference for evaluation of the effectiveness of Benford analysis, revealed a high percentage of incorrect data (66.5 %) which was dominated by falsified data. To characterize the degree of the data compliance with Benford's law, the mean absolute deviations of the frequency distributions of the first and first two significant digits deviations from the proper values and chi-squared statistics for the tenth powers of the standardized resistance, reactance, and resistance index values were assessed for each health center. A significant correlation was observed between the data deviation from Benford's law and the percentage of incorrect data as provided by the expert quality assessment algorithm (ρmax = 0.66 and 0.62 for the mean absolute deviations and χ2 statistics, respectively, based on the resistance value and the first significant digit). It is suggested that deviation of the BIA data from Benford's law serves as a sufficient, but not a necessary, condition for their contamination. For those health centers, in which most of the incorrect data were represented by multiple measurements of the same person under the guise of different ones, the data were in good agreement with Benford's law. If the structure of incorrect data was dominated by measurements of the calibration block, software emulations of BIA measurements and outliers, then the use of Benford's law made it possible to effectively rank health centers by the level of data authenticity.

ECNN: Intelligent Fault Diagnosis Method Using Efficient Convolutional Neural Network

With outstanding deep feature learning and nonlinear classification abilities, Convolutional Neural Networks (CNN) have been gradually applied to deal with various fault diagnosis tasks. Affected by variable working conditions and strong noises, the empirical datum always has different probability distributions, and then different data segments may have inconsistent contributions, so more attention should be assigned to the informative data segments. However, most of the CNN-based fault diagnosis methods still retain black-box characteristics, especially the lack of attention mechanisms and ignoring the special contributions of informative data segments. To address these problems, we propose a new intelligent fault diagnosis method comprised of an improved CNN model named Efficient Convolutional Neural Network (ECNN). The extensive view can cover the special characteristic periods, and the small view can locate the essential feature using Pyramidal Dilated Convolution (PDC). Consequently, the receptive field of the model can be greatly enlarged to capture the location information and excavate the remarkable informative data segments. Then, a novel residual network feature calibration and fusion (ResNet-FCF) block was designed, which uses local channel interactions and residual networks based on global channel interactions for weight-redistribution. Therefore, the corresponding channel weight is increased, which puts more attention on the information data segment. The ECNN model has achieved encouraging results in information extraction and feature channel allocation of the feature. Three experiments are used to test different diagnosis methods. The ECNN model achieves the highest average accuracy of fault diagnosis. The comparison results show that ECNN has strong domain adaptation ability, high stability, and superior diagnostic performance.