Performance improvement of fire detectors by means of gas sensors and neural networks

Nowadays, fire detection systems are used world wide in order to protect life and goods. However, at the present time, detectors show poor features with regard to detection speed and reliability. They respond to only a few fire parameters like smoke particles and they do not take into account other important fire parameters such as gaseous products. In this paper, we present a new multi-sensor detector consisting of a commercial optical fire detector, a temperature sensor and selected semiconductor metal oxide gas sensors. The use of a multi-sensor detector requires a more sophisticated algorithm than the simple threshold rule. The new algorithms are typically based on pattern recognition systems, consisting of a pre-processing unit, a feature extraction unit and a classification unit. The choice of suitable methods for the feature extraction and the classification is difficult. Most often, the classifier depends on the type of the extracted features. In this paper two methods for the feature extraction with their suitable classifiers are presented and compared. However the classification is based on neural networks.The first algorithm consists of (i) a pre-processing unit; and (ii) a FFT-based feature extraction unit to resolve characteristic fire signatures. For that purpose a moving window has been introduced and a composed signal has been generated from the different pre-processed sensor responses. The algorithm also consists of (iii) a classification unit with a Learning Vector Quantization (LVQ) neural network to classify the extracted features to fire, not fire, or disturbing event.The second algorithm consists of a pre-processing unit and feature extraction method based on the scaling of the quadratic mean value. For this kind of feature extraction a back-propagation neural network as a classifier has been chosen.An important improvement towards the use of commercial detectors has been achieved using both of the above-described algorithms. The neural networks with suitable feature extraction methods were able to detect test fires more quickly than the commercial optical fire detector without generating false alarms by disturbing events.     

Design and test methods for a video-based cargo fire verification system for commercial aircraft

The Cargo Fire Verification System was developed to address the problem of frequent false smoke alarms that are of particular concern in long range flights of passenger aircraft. The system uses low-cost CCD cameras operating in the near infra red range to directly detect fire and hotspots. In addition, LED illumination units are appropriately switched on and off, and the obtained images are analyzed to detect smoke. Fusion of image processing results with temperature and humidity readings allows reliable detection of true fires and elimination of false alarms due to fog and dust. It was necessary to create a suite of fire sensitivity and false alarm immunity tests applicable to these vision-based fire detection systems. This paper will concentrate on the design of the system, testing aspects, and test cell modifications.     

Experimental validation of a deep neural network—Sparse representation classification ensemble method

In the current study, a new pattern recognition‐based damage detection technique is developed using the frequency response function of the structure. Principal component analysis is employed as an authoritative feature extraction method for dimensional reduction of the measured frequency response function data and constructing distinct feature patterns. Subsequently, as a novel approach, an ensemble of 2 powerful classifiers containing deep neural networks and couple sparse coding classification is utilized for damage prediction of the structure because there is no individual optimal classifier for all the problems. Verification of the proposed method is evaluated by an aluminum beam experimental setup besides a numerical 3D finite element model of a truss bridge. Damage detection results elucidate that the ensemble method decisions are much more accurate compared with the individual classifier decision. The proposed ensemble method verifies to be a novel, robust, and powerful damage detection process.     

Using multivariate statistical methods to detect fires

Fire detectors must accurately detect fires, but they should not respond to false alarms. Contemporary smoke detectors sometimes cannot discriminate between smoke and odor sources. These detectors can also be slow in responding to smoldering fire sources. In this paper, a statistical approach for detecting fires based on fusing sensor signals from multiple sensors is presented. The multivariate statistical approach, called principal component analysis, is used to compress the sensor information down to a small number of variables that can be interpreted more easily than the raw sensor signals themselves. Experimental results presented here show that the proposed approach is more accurate than a conventional smoke alarm, particularly for early detection of smoldering fires. However, this new approach does not overcome the problem of false alarms. In spite of this current limitation, the method discussed holds great promise for future fire detection applications.     

Structural Health Monitoring via Measured Ritz Vectors Utilizing Artificial Neural Networks

Abstract:   A pattern recognition approach for structural health monitoring (SHM) is presented that uses damage-induced changes in Ritz vectors as the features to characterize the damage patterns defined by the corresponding locations and severity of damage. Unlike most other pattern recognition methods, an artificial neural network (ANN) technique is employed as a tool for systematically identifying the damage pattern corresponding to an observed feature. An important aspect of using an ANN is its design but this is usually skipped in the literature on ANN-based SHM. The design of an ANN has significant effects on both the training and performance of the ANN. As the multi-layer perceptron ANN model is adopted in this work, ANN design refers to the selection of the number of hidden layers and the number of neurons in each hidden layer. A design method based on a Bayesian probabilistic approach for model selection is proposed. The combination of the pattern recognition method and the Bayesian ANN design method forms a practical SHM methodology. A truss model is employed to demonstrate the proposed methodology.     

Acoustic emission detection of fatigue damage in cruciform welded joints

Weld seams are critical points for the initiation of fatigue cracks in steel structures subjected to cyclic loads. Semi-elliptical surface cracking at the toes of a fillet weld is not easily found when it is partially through the thickness and subcritical. In this study the acoustic emission (AE) method is used to detect crack propagation in cruciform fillet welded joints that are representative of typical fatigue sensitive details in steel bridge superstructures. The effect of geometry and fatigue load on the AE data is investigated by varying the width of the base plate and the stress ratio. AE data filtering based on load pattern, source location, and waveform feature analysis was implemented to minimize noise-induced AE signals and false indications due to wave reflections. AE time domain features such as amplitude (b-value), counts, signal strength, and absolute energy are employed to study the influence of geometry and fatigue load on the AE data.     

An Improved Probabilistic Approach for Fire Detection in Videos

This paper proposes an improved probabilistic approach using two improved feature representations. These features are color and motion. First, an improved probabilistic model for color-based fire detection is proposed, and candidate fire regions are generated from this model. Then, an improved motion feature is used for final decision. The performance of the proposed approach showed about 0.2758 accuracy in false positive rate, and 0.2636 accuracy in false negative rate on a benchmark fire video database, which represents a decrease of 46.6% in false positive rate, and a decrease of 52.1% in false negative rate compared to the probabilistic approach.     

基于MK-XGBoost的多传感器融合火灾识别技术

针对单一传感器预测中的漏报、误报的缺点,本文提出了一种基于MK-XGBoost的多传感器数据融合火灾识别技术.该技术通过采集受限空间的温度、烟雾质量浓度、CO体积分数,基于Mann-Kendall检验方法生成趋势因子,该因子与上升趋势呈正相关,后续将火灾数据与趋势因子作为特征,采用XG-Boost算法判断火灾是否发生....     

Feature Extraction for Traffic Incident Detection Using Wavelet Transform and Linear Discriminant Analysis

To eliminate false alarms, an effective traffic incident detection algorithm must be able to extract incident-related features from the traffic patterns. A robust feature-extraction algorithm also helps reduce the dimension of the input space for a neural network model without any significant loss of related traffic information, resulting in a substantial reduction in the network size, the effect of random traffic fluctuations, the number of required training samples, and the computational resources required to train the neural network. This article presents an effective traffic feature-extraction model using discrete wavelet transform (DWT) and linear discriminant analysis (LDA). The DWT is first applied to raw traffic data, and the finest resolution coefficients representing the random fluctuations of traffic are discarded. Next, LDA is employed to the filtered signal for further feature extraction and reducing the dimensionality of the problem. The results of LDA are used as input to a neural network model for traffic incident detection.     

北京市延庆硅化木园区典型岩石矿物图像特征提取与分析

北京市延庆硅化木园区位于中朝准地台内蒙地轴与燕山台褶带两个Ⅱ级构造单元衔接部位,地层岩性组合在北京地区具有一定的典型性和代表性。对研究区内岩石矿物特征的提取与分析,进而为北京地区今后开展岩性识别相关工作具有一定的参考意义。本研究基于Visual C++软件平台,应用数字图像处理技术开展颜色、灰度、纹理、投影、频域等多种适用于典型岩石矿物图像的特征提取方法研究,进行系统的参数值计算及仿真分析。通过优化选择确定最优特征量及最大判据值,实现了多参数特征有效集成,初步建立了综合、全面的特征提取系统技术体系,有助于提升地质调查工程中岩石矿物识别的精度。     

Large-scale building damage assessment using a novel hierarchical transformer architecture on satellite images

This paper presents damage assessment using a hierarchical transformer architecture (DAHiTrA), a novel deep-learning model with hierarchical transformers to classify building damages based on satellite images in the aftermath of natural disasters. Satellite imagery provides real-time and high-coverage information and offers opportunities to inform large-scale postdisaster building damage assessment, which is critical for rapid emergency response. In this work, a novel transformer-based network is proposed for assessing building damage. This network leverages hierarchical spatial features of multiple resolutions and captures the temporal differences in the feature domain after applying a transformer encoder to the spatial features. The proposed network achieves state-of-the-art performance when tested on a large-scale disaster damage data set (xBD) for building localization and damage classification, as well as on LEVIR-CD data set for change detection tasks. In addition, this work introduces a new high-resolution satellite imagery data set, Ida-BD (related to 2021 Hurricane Ida in Louisiana) for domain adaptation. Further, it demonstrates an approach of using this data set by adapting the model with limited fine-tuning and hence applying the model to newly damaged areas with scarce data.     

Factorial design–machine learning approach for predicting incident durations

This research proposes a hybrid approach for predicting incident duration that integrates the salient features of both factorial design of experiments (DOE) and machine learning (ML). This study compares DOE with another widely used technique, forward sequential feature selection (FSFS). Moreover, to confirm the effectiveness and robustness of the proposed approach, multiple ML techniques are employed, including linear regression, decision trees, support vector machines, ensemble trees, Gaussian process regression, and artificial neural networks. The study results are validated using data from the Houston TranStar incidents archive with over 90,000 records. The accuracy of the developed predictive models is compared based on multiple techniques (i.e., no feature selection–ML, FSFS–ML, and DOE–ML). The results revealed that the significant factors affecting incident duration identified by both DOE and FSFS include the type of vehicles involved, type of lanes affected, number of vehicles involved, number of emergency responses dispatched, incident severity level, and day of the week. The comparative results of the different feature selection and modeling approaches revealed that the hybrid DOE–ML approach outperformed the other tested analysis approaches. The best-performing model under the DOE–ML approach was the SVM with cubic kernel model. It reduced the modeling time by 83.8% while increasing the prediction error by merely 0.02%, which is not significant. Therefore, the prediction accuracy could be slightly downgraded in return for a substantial reduction in the number of variables utilized, resulting in substantial savings in the modeling time and required dataset.     

Reference-Free Damage Classification Based on Cluster Analysis

Abstract:   Fiber-reinforced polymer (FRP) composite materials have been widely used for retrofitting civil infrastructure systems. The ultimate goal of this study was to develop an in-site non-destructive testing (NDT) technique that can continuously and autonomously inspect the bonding condition between a carbon FRP (CFRP) layer and a host reinforced concrete (RC) structure, when the CRFP layer is used for strengthening the RC structure. The uniqueness of this reference-free NDT is two-fold: First, features, which are sensitive to CFRP debonding but insensitive to operational and environmental variations of the structure, have been extracted only from current data without direct comparison with previously obtained baseline data. Second, damage classification is performed instantaneously without relying on predetermined decision boundaries. The extraction of the reference-free features is accomplished based on the concept of time reversal acoustics, and the instantaneous decision-making is achieved using cluster analysis. Monotonic and fatigue load tests of large-scale CFRP-strengthened RC beams are conducted to demonstrate the potential of the proposed reference-free debonding monitoring technique. Based on the experimental studies, it has been shown that the proposed reference-free NDT technique may minimize false alarms of debonding and unnecessary data interpretation by end users.     

Two Novel Approaches to Reduce False Alarm Due to Non‐Earthquake Events for On‐Site Earthquake Early Warning System

An on‐site earthquake early warning system (EEWS) can provide more lead‐time at regions that are close to the epicenter of an earthquake because only seismic information of a target site is required. Instead of leveraging the information of several stations, the on‐site system extracts some P‐wave features from the first few seconds of vertical ground acceleration of a single station. It then predicts the intensity of the forthcoming earthquake at the same station according to these features. However, the system may be triggered by some vibration signals that are not caused by an earthquake or by interference from electronic signals, which may consequently result in a false alarm at the station. Thus, this study proposes two approaches to distinguish the vibration signals caused by non‐earthquake events from those caused by earthquake events based on support vector classification (SVC) and singular spectrum analysis (SSA). In the first approach (Approach I), the fast Fourier transform algorithm and the established SVC model are employed to classify the vibration signals. In the second approach (Approach II), a SSA criterion is added to Approach I for the purpose of identifying earthquake events that are classified as non‐earthquake events by the SVC model with increased accuracy. Both approaches are verified by using data collected from the Taiwan Strong Motion Instrumentation Program and EEW stations of the National Center for Research on Earthquake Engineering. The results indicate that both of the proposed approaches effectively reduce the possibility of false alarms caused by an unknown vibration event.     

基于EMD_SVD的矿山微震与爆破信号特征提取及分类方法

针对矿山微震与爆破信号难以识别的问题,提出了基于经验模态分解(EMD)和奇异值分解(SVD)的矿山信号特征提取及分类方法。首先对微震与爆破信号进行EMD分解,再借助相关系数和方差贡献率筛选得到主要本征模态分量为IMF1~IMF6,进而利用SVD计算主要本征模态分量构成矩阵的奇异值σ_i(i=1,2,...,6),最后应用支持向量机(SVM)对用沙坝矿微震与爆破信号进行分类。结果表明:微震与爆破信号的奇异值σ_1,σ_2和σ_3差异较大,且σ_1(28)7.5作为识别分界值时准确率达到了88.25%;SVM法识别效果优于BP神经网络法、Bayes法和单一奇异值分界值法,且SVM法准确率达到了93.0%。由此,该方法可为矿山微震与爆破信号特征提取和分类提供一种新方法。     

A texture‐Based Video Processing Methodology Using Bayesian Data Fusion for Autonomous Crack Detection on Metallic Surfaces

Regular inspection of the components of nuclear power plants is important to improve their resilience. However, current inspection practices are time consuming, tedious, and subjective: they involve an operator manually locating cracks in metallic surfaces in the plant by watching videos. At the same time, prevalent automatic crack detection algorithms may not detect cracks in metallic surfaces because these are typically very small and have low contrast. Moreover, the existences of scratches, welds, and grind marks lead to a large number of false positives when state‐of‐the‐art vision‐based crack detection algorithms are used. In this study, a novel crack detection approach is proposed based on local binary patterns (LBP), support vector machine (SVM), and Bayesian decision theory. The proposed method aggregates the information obtained from different video frames to enhance the robustness and reliability of detection. The performance of the proposed approach is assessed by using several inspection videos. The results indicate that it is accurate and robust in cases where state‐of‐the‐art crack detection approaches fail. The experiments show that Bayesian data fusion improves the hit rate by 20% and the hit rate achieves 85% with only one false positive per frame.     

Machine learning based novelty detection using modal analysis

Among many structural assessment methods, the change of modal characteristics is considered a well‐accepted damage detection method. However, the presence of environmental or operational variations may pollute the baseline and prevent a dependable assessment of the change. In recent years, the use of machine learning algorithms gained interest within structural health community, especially due to their ability and success in the elimination of ambient uncertainty. This paper proposes an end‐to‐end architecture to detect damage reliably by employing machine learning algorithms. The proposed approach streamlines (a) collection of structural response data, (b) modal analysis using system identification, (c) learning model, and (d) novelty detection. The proposed system aims to extract latent features of accessible modal parameters such as natural frequencies and mode shapes measured at undamaged target structure under temperature uncertainty and to reconstruct a new representation of these features that is similar to the original using well‐established machine learning methods for damage detection. The deviation between measured and reconstructed parameters, also known as novelty index, is the essential information for detecting critical changes in the system. The approach is evaluated by analyzing the structural response data obtained from finite element models and experimental structures. For the machine learning component of the approach, both principal component analysis (PCA) and autoencoder (AE) are examined. While mode shapes are known to be a well‐researched damage indicator in the literature, to our best knowledge, this research is the first time that unsupervised machine learning is applied using PCA and AE to utilize mode shapes in addition to natural frequencies for effective damage detection. The detection performance of this pipeline is compared to a similar approach where its learning model does not utilize mode shapes. The results demonstrate that the effectiveness of the damage detection under temperature variability improves significantly when mode shapes are used in the training of learning algorithm. Especially for small damages, the proposed algorithm performs better in discriminating system changes.     

Wildfire smoke detection based on local extremal region segmentation and surveillance

A novel video-based method is proposed for long-distance wildfire smoke detection. Since the long-distance wildfire smoke usually moves slowly and lacks salient features in the video, the detection is still a challenging problem. Unlike many traditional video-based methods that usually rely on the smoke color or motion for initial smoke region segmentation, we use the Maximally Stable Extremal Region (MSER) detection method to extract local extremal regions of the smoke. This makes the initial segmentation of possible smoke region less dependent on the motion and color information. Potential smoke regions are then selected from all the possible regions by using some static visual features of the smoke, helping to eliminate the non-smoke regions as many as possible. Once a potential smoke region is found, we keep tracking it by searching the best-matched extremal regions in the subsequent frames. At the same time, the propagating motions of the potential smoke region are monitored based on a novel cumulated region approach, which can be effectively used to identify the distinctive expanding and rising motions of smoke. This approach can also make the final smoke motion identification insensitive to image shaking. It was proved that the proposed method is able to reliably detect the long-distance wildfire smoke and simultaneously produce very few false alarms in actual applications.     

Long-term bridge health monitoring and performance assessment based on a Bayesian approach

This study presents a damage detection approach for the long-term health monitoring of bridge structures. The Bayesian approach comprising both Bayesian regression and Bayesian hypothesis testing is proposed to detect the structural changes in an in-service seven-span steel plate girder bridge with Gerber system. Both temperature and vehicle weight effects are accounted in the analysis. The acceleration responses at four points of the bridge span are utilised in this investigation. The data covering three different time periods are used in the bridge health monitoring (BHM). Regression analyses showed that the autoregressive exogenous model considering both temperature and vehicle weight effects has the best performance. The Bayesian factor is found to be a sensitive damage indicator in the BHM. The Bayesian approach can provide updated information in the real-time monitoring of bridge structures. The information provided from the Bayesian approach is convenient and easy to handle compared to the traditional approaches. The applicability of this approach is also validated in a case study where artificially generated damage data is added to the observation data.     

基于神经网络的桥梁损伤识别研究综述

对人工神经网络的基本原理、特点以及与损伤识别的关系作了简要介绍,并重点介绍了损伤识别中常用的BP 神经网络的原理及其改进方法,以及国内外在基于神经网络的桥梁损伤识别应用方面的主要研究成果,最后对神经网络在桥梁损伤识别中的发展和应用作了展望。