Inspection Algorithm Research Articles

A novel deep learning algorithm applied to machine vision inspection for surface defects of injection moulded products

In modern industry, the surface defect inspection of injection moulded products is crucial for controlling product quality and optimising the manufacturing process. With the development of optical measurement and computer technology, machine vision inspection methods have been widely adopted instead of manual inspection. However, current machine vision inspection methods are difficult to simultaneously ensure the accuracy and efficiency of surface defect inspection of injection moulded products. Considering this problem, a novel deep learning algorithm applied to machine vision inspection for surface defects of injection moulded products is proposed. To train and evaluate the proposed deep learning algorithm, an image acquisition platform is established and the dataset of surface defects in moulded products is obtained. In the proposed deep learning algorithm, reparameterisation-based convolution modules are employed for feature extraction and feature fusion. A median iterative clustering algorithm based on hierarchical clustering initialisation is proposed to obtain prior anchors that are highly matched with the actual distribution of defect sizes. A novel focus-entire union over covering loss function is utilised for bounding box regression. On these bases, the proposed deep learning algorithm applied to machine vision inspection is evaluated on the dataset of surface defects in moulded products. The experimental results indicate that compared to the traditional inspection algorithms and other deep learning algorithms currently used in machine vision inspection, the proposed deep learning algorithm exhibits superior inspection accuracy and inspection efficiency on the acquired dataset. The inspection precision reaches 0.964, the inspection recall reaches 0.955, and the inference time for each subgraph is only 6.1 ms, confirming its effectiveness.

Efficient Defect Detection of Rotating Goods under the Background of Intelligent Retail.

Dynamic visual vending machines are rapidly growing in popularity, offering convenience and speed to customers. However, there is a prevalent issue with consumers damaging goods and then returning them to the machine, severely affecting business interests. This paper addresses the issue from the standpoint of defect detection. Although existing industrial defect detection algorithms, such as PatchCore, perform well, they face challenges, including handling goods in various orientations, detection speeds that do not meet real-time monitoring requirements, and complex backgrounds that hinder detection accuracy. These challenges hinder their application in dynamic vending environments. It is crucial to note that efficient visual features play a vital role in memory banks, yet current memory repositories for industrial inspection algorithms do not adequately address the problem of location-specific feature redundancy. To tackle these issues, this paper introduces a novel defect detection algorithm for goods using adaptive subsampling and partitioned memory banks. Firstly, Grad-CAM is utilized to extract deep features, which, in combination with shallow features, mitigate the impact of complex backgrounds on detection accuracy. Next, graph convolutional networks extract rotationally invariant features. The adaptive subsampling partitioned memory bank is then employed to store features of non-defective goods, which reduces memory consumption and enhances training speed. Experimental results on the MVTec AD dataset demonstrate that the proposed algorithm achieves a marked improvement in detection speed while maintaining accuracy that is comparable to state-of-the-art models.

Rapid damage reconstruction imaging of composite plates using non-contact air-coupled Lamb waves

Air-coupled ultrasonic Lamb wave damage imaging is a method for rapid, non-contact damage detection of plate specimens. However, the current technique based on Lamb waves has limitations regarding accuracy and quality of imaging. To address this issue, an improved bilateral boundary-enhanced reconstruction algorithm for the probabilistic inspection of defects (BBERAPID) is proposed. This algorithm optimizes the probability distribution near the damage boundary to enhance the accuracy of damage imaging. The skewed bilateral boundary distribution algorithm can improve the probability distribution of edges on both sides of the damage and alleviate the problem of abnormal boundary expansion. In addition, a new damage index (DI) calculation method is proposed to accelerate the DI trend near the damage. The new DI can improve the fineness of the damage boundary, thus better distinguish the healthy area from the damage area. Traditional DI has the problem of abnormal distribution at the damage boundary, which BBERAPID solves. The probability distribution near the damage is optimized to improve the quality and accuracy of damage imaging. Experimental studies are performed using composite plates to validate the proposed method further. The effectiveness and accuracy of the proposed BBERAPID algorithm for detecting damages in composite plates are verified by comparing it with existing imaging methods through experimental validation. The proposed non-contact air-coupled Lamb wave damage imaging method is suitable for rapidly and accurately characterizing defects in composite plates.

Multi-Objective Optimization-Based Planning Algorithm for Efficient Visual Inspection

Multi-Objective Optimization-Based Planning Algorithm for Efficient Visual Inspection

Research on multiple damage localisation based on fusion of the Lamb wave ellipse algorithm and RAPID algorithm

Current damage localisation methods often require many sensors and complex signal processing methods. This paper proposes a fusion algorithm based on elliptical localisation and the reconstruction algorithm for probabilistic inspection of damage (RAPID) to locate and image multiple damages. Experimental verification of the damage algorithm was conducted. An ultrasonic probe was used to excite Lamb signals on an aluminium alloy plate, the ultrasonic response signals at different positions within the plate under multiple damages were measured and the constructed algorithm was employed to image the damage location. In the experiment, this method improved localisation efficiency by excluding invalid sensing paths in the sensor network, saving 31.32% of computational time. When some sensors in the sensor network were damaged, this algorithm ensured a positioning accuracy with a positioning error of 5.83 mm. Finally, the algorithm was used to locate multiple damages in the sensor network and the results showed the good robustness of the algorithm.

Toward Real Scenery: A Lightweight Tomato Growth Inspection Algorithm for Leaf Disease Detection and Fruit Counting

The deployment of intelligent surveillance systems to monitor tomato plant growth poses substantial challenges due to the dynamic nature of disease patterns and the complexity of environmental conditions such as background and lighting. In this study, an integrated cascade framework that synergizes detectors and trackers was introduced for the simultaneous identification of tomato leaf diseases and fruit counting. We applied an autonomous robot with smartphone camera to collect images for leaf disease and fruits in greenhouses. Further, we improved the deep learning network YOLO-TGI by incorporating Ghost and CBAM modules, which was trained and tested in conjunction with premier lightweight detection models like YOLOX and NanoDet in evaluating leaf health conditions. For the cascading with various base detectors, we integrated state-of-the-art trackers such as Byte-Track, Motpy, and FairMot to enable fruit counting in video streams. Experimental results indicated that the combination of YOLO-TGI and Byte-Track achieved the most robust performance. Particularly, YOLO-TGI-N emerged as the model with the least computational demands, registering the lowest FLOPs at 2.05 G and checkpoint weights at 3.7 M, while still maintaining a mAP of 0.72 for leaf disease detection. Regarding the fruit counting, the combination of YOLO-TGI-S and Byte-Track achieved the best R 2 of 0.93 and the lowest RMSE of 9.17, boasting an inference speed that doubles that of the YOLOX series, and is 2.5 times faster than the NanoDet series. The developed network framework is a potential solution for researchers facilitating the deployment of similar surveillance models for a broad spectrum of fruit and vegetable crops.

CAD-based Autonomous Vision Inspection Systems

Automated industrial Visual Inspection Systems (VIS) are typically customized for specific applications, limiting their flexibility. They are characterized by a demanding setup, high capital investments, and significant knowledge barriers. In this paper, we propose an alternative architecture for the visual inspection of 3D printed parts or complex assemblies using a robotic arm equipped with hand-eye sensors and controllable lighting system. The core of the proposed Flexible Vision Inspection System (FVIS) is the self-extraction of 3D text annotations from STandard for the Exchange of Product model (STEP) AP242 files. The system self-selects and parametrizes the most suitable inspection algorithm, including lighting settings. Additionally, it autonomously performs self-localization, self-referencing of physical products, and self-planning of robot inspection path based on CAD information. This framework, characterized by self-X, cost-effective, non-invasive, and plug-and-play architecture has the potential to disrupt the business model of vision inspection, enabling an as-a-service solution aligned with the next generation of flexible manufacturing.

SCALABILITY IMPROVEMENT OF ACTIVE PROBING FOR FAILURE DETECTION IN LARGE-SCALE SOFTWARE DEFINED NETWORKS USING COMMUNITY DETECTION ALGORITHMS

Software-Defined Networking (SDN) is a networking approach that separates the network's control plane and data plane to provide better network control. The nature of SDN requires an essential focus on its reliability and availability, which includes failure detection aspects. Several techniques have been proposed to enhance SDN failure detection, including passive and active approaches. However, the Route Inspection (RI) algorithm, an example of the active probing method in active detection, has scalability issues when deployed in large-scale networks as it generates inefficient and lengthy probe paths. To address this, we propose using community detection (CD) algorithms to improve the probe path generation of the RI algorithm, resulting in the 'Failure Detection using Community Detection and probing (FDCD)' framework. With the aim to reduce travel paths, the proposed framework minimizes the network topology using CD algorithms to reduce the topology size into smaller communities. Then, the RI algorithm calculates the probe path in the communities instead of the single large topology, thus improving the efficiency of the failure detection mechanism. This paper adopts three community detection algorithms: Louvain, Label Propagation, and Girvan-Newman. We evaluate the framework using three well-known network topologies, COST266, PIORO40, and GERMANY50, in Mininet to observe the framework's performance. The results indicate that the proposed approach improves the algorithm's scalability against its baseline approach, improving the average probe round trip time by 66.87% and the average path installation time by 2.67%.

Continuous laser-line scanning thermography with data-processing algorithm for rapid and accurate defect inspection

As an emerging nondestructive evaluation method, line-scanning thermography is efficient for defect detection and location. In this study, continuous laser-line scanning thermography combined with a corresponding data-processing algorithm is developed to realize highly accurate and highly efficient nondestructive evaluation of large structures. First, the configuration of continuous laser-line-scanning thermography is investigated, and a data-processing algorithm is established using the visual reconstruction method and the distortion correction of captured thermal sequence images. Subsequently, experiments are conducted on samples with defects at different scanning speeds to verify the stability and accuracy of the proposed method. The results indicate that the continuous laser-line scanning thermography is robust for defect detection at different scanning speeds. Highly accurate measurements of defects can be obtained under information redundancy or absence when the sampling frequency does not exactly match the scanning speed. Quantitative experimental comparison shows that the relative error of the defect measurement reduces significantly after the distortion of the initial reconstructed thermal image is corrected. Finally, the effects of the scanning velocity and selected pixel line on the defect detectability are investigated for image reconstruction. The proposed approach is shown to be promising for the continuous nondestructive evaluation of large structures in terms of accuracy and efficiency.

Detection of surface defect on flexible printed circuit via guided box improvement in GA-Faster-RCNN network.

Industrial defect detection is a critical aspect of production. Traditional industrial inspection algorithms often face challenges with low detection accuracy. In recent years, the adoption of deep learning algorithms, particularly Convolutional Neural Networks (CNNs), has shown remarkable success in the field of computer vision. Our research primarily focused on developing a defect detection algorithm for the surface of Flexible Printed Circuit (FPC) boards. To address the challenges of detecting small objects and objects with extreme aspect ratios in FPC defect detection for surface, we proposed a guided box improvement approach based on the GA-Faster-RCNN network. This approach involves refining bounding box predictions to enhance the precision and efficiency of defect detection in Faster-RCNN network. Through experiments, we verified that our designed GA-Faster-RCNN network achieved an impressive accuracy rate of 91.1%, representing an 8.5% improvement in detection accuracy compared to the baseline model.

A privacy-preserving federated learning algorithm for intelligent inspection in pumped storage power station

A privacy-preserving federated learning algorithm for intelligent inspection in pumped storage power station

Investigation of Different Features for Baseline-Free RAPID Damage-Imaging Algorithm Using Guided Waves Applied to Metallic and Composite Plates

In guided-wave-based damage-imaging algorithms, damage reconstruction typically involves comparing the signals with and without a defect. However, in many cases, defect-free data may not be available. Therefore, in this study, baseline and baseline-free approaches were used for damage imaging, exploiting not only the amplitude of the signal as the feature but also five additional features, namely, the amplitude of the sparse signal after deconvolution, the amplitude of the coefficients at the excitation frequency from the re-assigned short-time Fourier transform, the time of flight determined from cross-correlation, kurtosis in the time domain, and kurtosis in the frequency domain. For this study, three different plates with different types of defects were considered: a metallic plate with a notch-type artificial defect, a pultruded type of composite plate with an impact defect, and a laminate composite plate with plexiglass serving as an added mass damper artificial defect. The Reconstruction Algorithm for Probabilistic Inspection of Damage (the RAPID algorithm) was used to characterize the defects on the three plates, and the defect parameters were then quantified by creating an ellipse after thresholding.

Visual analytics and intelligent reasoning for smart manufacturing defect detection and judgement: A meta-learning approach with knowledge graph embedding case-based reasoning

Visual inspection is an important procedure in manufacturing industries to guarantee product quality and reduce costs of poor quality. Several challenges exist for manufacturing inspection automation, including the false positive and false negative trade-off, autonomous decision-making, and inspection algorithm selection. This paper addresses these challenges by proposing a two-level visual analytics and intelligent reasoning system, where machine vision identifies and locates potential defects and convolutional neural network models further classify them into acceptable and defective features. Symbolic reasoning techniques are applied for autonomous decision making on part-level judgement to enhance the system reasoning capability, including part judgement (accept, reject, rework) and rework operation assignment. Furthermore, a case-based reasoning approach for meta-learning of inspection algorithm selection is proposed, which uses knowledge graph embedding (TransR embedding) to embed cases into a two-dimensional space for case similarity estimation and case retrieval. A case of smart assembly inspection is presented to validate the feasibility of the proposed system.

Baseline-Free Damage Imaging of Composite Lap Joint via Parallel Array of Piezoelectric Sensors.

This paper presents a baseline-free damage imaging technique using a parallel array of piezoelectric sensors and a control board that facilitates custom combinations of sensor selection. This technique incorporates an imaging algorithm that uses parallel beams for generation and reception of ultrasonic guided waves in a pitch-catch configuration. A baseline-free reconstruction algorithm for probabilistic inspection of defects (RAPID) algorithm is adopted. The proposed RAPID method replaces the conventional approach of using signal difference coefficients with the maximum signal envelope as a damage index, ensuring independence from baseline data. Additionally, conversely to the conventional RAPID algorithm which uses all possible sensor combinations, an innovative selection of combinations is proposed to mitigate attenuation effects. The proposed method is designed for the inspection of lap joints. Experimental measurements were carried out on a composite lap joint, which featured two dissimilar-sized disbonds positioned at the lap joint's borderline. A 2D correlation coefficient was used to quantitatively determine the similarity between the obtained images and a reference image with correct defect shapes and locations. The results demonstrate the effectiveness of the proposed damage imaging method in detecting both defects. Additionally, parametric studies were conducted to illustrate how various parameters influence the accuracy of the obtained imaging results.

Advanced Dielectric Anomaly Detection for Non-Destructive Fruit Inspection Using Electromagnetic Microwave Technique

This paper presents a comprehensive investigation into the application of an electromagnetic microwave technique combined with the dielectric anomaly approach algorithm for the non-destructive inspection of fruits. The proposed antenna configuration comprises eight elements arranged in a circular array, enabling the collection of signals scattered by objects, specifically fruits, placed in their path. The collected data undergoes a series of processing steps, including Fast Fourier Transform, covariance matrix estimation, eigenvalue and eigenvector computation, and spatial spectrum construction. The dielectric anomaly algorithm is then applied to detect defects in the fruits. The study covers five different types of fruits, both healthy and defective, and gathers essential dielectric properties. Furthermore, a novel hybrid IQR method is introduced for outlier detection in the dielectric data. The results demonstrate the effectiveness of the proposed methodology in providing valuable insights into the internal structures of fruits and detecting anomalies, contributing to the enhancement of quality assessment in fruit inspection processes.

A census of quality-controlled Biogeochemical-Argo float measurements

Biogeochemical- (BGC-) Argo aims to deploy and maintain a global array of autonomous profiling floats to monitor ocean biogeochemistry. With over 250,000 profiles collected so far, the BGC-Argo network is rapidly expanding toward the target of a sustained fleet of 1,000 floats. These floats prioritize the measurement of six key properties: oxygen, nitrate, pH, chlorophyll-a, suspended particles, and downwelling light. To assess the current biogeochemical state of the ocean, its variability, and trends with confidence, it is crucial to quality control these measurements. Accordingly, BGC-Argo maintains a quality control system using manual inspection and parameter-specific algorithms for flagging and adjusting data. In this study, we provide a census of the quantity and quality of measurements from BGC-Argo based on their quality flagging system. The purpose of this census is to assess the current status of the array in terms of data quality, how data quality has changed over time, and to provide a better understanding of the quality-controlled data to current and future users. Alongside increasing profile numbers and spatial coverage, we report that for most parameters between 80 and 95% of the profiles collected so far contain high-quality BGC data, with an exception for pH. The quality of pH profiles has seen a large improvement in the last five years and is on track to match the data quality of other BGC parameters. We highlight how BGC-Argo is improving and discuss strategies to increase the quality and quantity of BGC profiles available to users. This census shows that tracking percentages of high-quality data through time is useful for monitoring float sensor technology and helpful for ensuring the long-term success of BGC-Argo.

Slope displacement detection in construction: An automated management algorithm for disaster prevention

In construction sites, embankment slope failure has led to catastrophic events, necessitating advanced monitoring for disaster prevention. This study presents a novel Slope Displacement Inspection and Management Algorithm (SDIMA) that utilizes Global Navigation Satellite System (GNSS) data to detect significant slope displacements automatically. SDIMA analyzes GNSS data to discern displacements in various directions (North-South, East-West, Vertical) and separate actual change points from measurement errors by employing a robust mathematical model called Displacement Detector. It successfully detected change points with displacements greater than two times the samples' standard deviation, and thresholds were set for displacements ≥ 25 mm. A comparative analysis showed promising efficiency in North-South detection, while certain limitations were identified. The proposed SDIMA outperforms existing models, achieving a Mean Absolute Error (MAE) of 3.157 mm, Root Mean Square Error (RMSE) of 4.754 mm, Mean Absolute Scaled Error (MASE) of 0.294, and an average processing time of 0.578 min, making it a highly efficient and accurate expert system. The innovation of SDIMA lies in its ability to provide an automated decision support tool for site managers, laying the foundation for a low-cost anomaly detection system for slope failure disaster prevention. The algorithm's proficiency and its potential to integrate with existing practices herald a significant advancement in construction management and safety.

Transformer-based visual inspection algorithm for surface defects

Transformer-based visual inspection algorithm for surface defects

Multi-type damage localization using the scattering coefficient-based RAPID algorithm with damage indexes separation and imaging fusion

Guided waves-based structural health monitoring (SHM) methods have potential for practical applications, since they are sensitive to small damages and are able to realize large area monitoring. Among these methods, the Reconstruction Algorithm for Probabilistic Inspection (RAPID), using a Piezoelectric transducer (PZT) sensor array, is one of the most widely used imaging algorithms to perform active damage monitoring and localization. However, since the sensing paths are distributed inside the sensor array with the non-uniform density, the RAPID algorithm can only localize damage when it is occurring inside of the array. If the damage occurs outside of the array or both inside and outside of the array, that is, multi-type damage, the performance of RAPID algorithm would not be satisfactory. In this paper, a scattering coefficient-based RAPID algorithm with damage indexes separation and imaging fusion is proposed. The amplitude of damage scattered signal at the corresponding time of fight is adopted as the weight in the probability distribution function, and damage indexes are then classified into two categories in the RAPID algorithm for the inside and outside damage localization respectively. Finally, an experiment on the complex composite plate, with the center large hole and surrounding bolt holes, is carried out to verify this proposed method. Experimental results show that this method can realize multi-type damage localization with errors less than 40 mm.

Enhancing Lamb wave-based damage diagnosis in composite materials using a pseudo-damage boosted convolutional neural network approach

Damage diagnosis of thin-walled structures has been successfully performed through methods based on tomography and machine learning-driven methods. According to traditional approaches, diagnostic signals are excited and sensed on the structure through a permanently installed network of sensors and are processed to obtain information about the damage. Good performance characterizes methods that process Lamb waves, which are described by long propagation distances and high sensitivity to anomalies. Most of the methods require extracting damage-sensitive features from the diagnostic signals to drive the damage diagnosis task. However, this process can lead to loss of information, and the choice of the specific feature to extract may introduce biases that hamper damage diagnosis. Furthermore, traditional approaches do not perform well when composites are considered, due to the anisotropy, inhomogeneity, and complex damage mechanisms shown by this type of material. To boost the performance of methods for damage diagnosis of composite plates, this work proposes a convolutional neural network (CNN)-based algorithm that localizes damage by processing Lamb waves. Different from other methods, the proposed method does not require extracting features from the acquired signals and allows localizing damage through the regression approach. The method was tested against experimental observations of Lamb waves propagating in two composite panels and in a hybrid panel, and the performance of two different sensor arrays was investigated. The pseudo-damage approach was used to generate large enough datasets for training the CNNs, and the performance of the framework was evaluated by localizing pseudo-damage and real damage determined by low-velocity impacts. The CNN-driven method accurately localized damage in all the considered scenarios, and it also outperformed traditional damage indices-based approaches, such as the reconstruction algorithm for probabilistic inspection of defects.