Automated diagnosis of sewer pipe defects based on machine learning approaches

In sewage rehabilitation planning, closed circuit television (CCTV) systems are the widely used inspection tools in assessing sewage structural conditions for non man entry pipes. Currently, the assessment of sewage structural conditions by manually interpretation on CCTV images seems inefficient, especially for several thousands of frames in one inspection plan. Also, the assessment work significantly involves engineers’ eye sight and professional experience. With a purpose of assisting general staffs in diagnosing pipe defects on CCTV inspection images, a diagnostic system by applying machine learning approaches is proposed in this paper. This research was first to use image process techniques, including wavelet transform and computation of co-occurrence matrices, for describing the textures of the pipe defects. Then, three neural network approaches, back-propagation neural network (BPN), radial basis network (RBN), and support vector machine (SVM), were adopted to classify pipe defect patterns, and their performances were compared and discussed. The diagnostic system of pipe defects was applied to a sewer system in the 9th district, Taichung City which is the largest city in middle Taiwan. The result shows that the diagnosis accuracy of 60% derived by SVM is the best and also better than the diagnosis accuracy of 57.4% derived by a Bayesian classifier.     

Model falsification diagnosis and sensor placement for leak detection in pressurized pipe networks

Pressurized pipe networks used for fresh-water distribution can take advantage of recent advances in sensing technologies and data-interpretation to evaluate their performance. In this paper, a leak-detection and a sensor placement methodology are proposed based on leak-scenario falsification. The approach includes modeling and measurement uncertainties during the leak detection process. The performance of the methodology proposed is tested on a full-scale water distribution network using simulated data. Findings indicate that when monitoring the flow velocity for 14 pipes over the entire network (295 pipes) leaks are circumscribed within a few potential locations. The case-study shows that a good detectability is expected for leaks of 50 L/min or more. A study of measurement configurations shows that smaller leak levels could also be detected if additional pipes are instrumented.     

An Empirical Evaluation of Probability Estimation with Neural Networks

Recent interest in neural networks by researchers across a wide spectrum of disciplines has provided convincing evidence of their ability to address classification problems. In this article, we consider the issue of evaluating the predictive capability of neural networks when the output values are to be treated as probabilities. We propose the use of a variant of a chi-square statistic, based on the Hosmer–Lemeshow statistic from logistic regression, to measure the goodness-of-fit of neural network models for two-group membership problems. Through experimentation with a large real-world database, we demonstrate the application of this statistic, and examine the effects of varying the number of nodes in the hidden layer on its value. Our empirical results suggest that this statistic can be very useful in identifying significant differences in the probability estimation accuracy of neural network models.       

Neural networks deterioration models for serviceability condition of buried stormwater pipes

Buried stormwater pipe networks play a key role in surface drainage systems for urban areas of Australia. The pipe networks are designed to convey water from rainfall and surface runoff only and do not transport sewage. The deterioration of stormwater pipes is commonly graded into structural and serviceability condition using CCTV inspection data in order to recognize two different deterioration processes and consequences. This study investigated the application of neural networks modelling (NNM) in predicting serviceability deterioration that is associated with reductions of pipe diameter until a complete blockage. The outcomes of the NNM are predictive serviceability condition for individual pipes, which is essential for planning proactive maintenance programs, and ranking of pipe factors that potentially contribute to the serviceability deterioration. In this study the Bayesian weight estimation using Markov Chain Monte Carlo simulation was used for calibrating the NNM on a case study in order to account for the uncertainty often encountered in NNM's calibration using conventional back-propagation weight estimation. The performance and the ranked factors obtained from the NNM were also compared against a classical model using multiple discrimination analysis (MDA). The results showed that the predictive performance of the NNM using Bayesian weight estimation is better than that of the NNM using conventional backpropagation and MDA model. Furthermore, among nine input factors, ‘pipe age’ and ‘location’ appeared insignificant whilst ‘pipe size’, ‘slope’, ‘the number of trees’ and ‘climatic condition’ were found consistently important over both models for serviceability deterioration process. The remaining three factors namely, ‘structure’, ‘soil’ and ‘buried depth’ might be redundant factors. A better and more consistent data collection regime may help to improve the predictive performance of the NNM and identify the significant factors.     

Prediction of compressive strength of concretes containing metakaolin and silica fume by artificial neural networks

Neural networks have recently been widely used to model some of the human activities in many areas of civil engineering applications. In the present paper, the models in artificial neural networks (ANN) for predicting compressive strength of concretes containing metakaolin and silica fume have been developed at the age of 1, 3, 7, 28, 56, 90 and 180 days. For purpose of building these models, training and testing using the available experimental results for 195 specimens produced with 33 different mixture proportions were gathered from the technical literature. The data used in the multilayer feed forward neural networks models are arranged in a format of eight input parameters that cover the age of specimen, cement, metakaolin (MK), silica fume (SF), water, sand, aggregate and superplasticizer. According to these input parameters, in the multilayer feed forward neural networks models are predicted the compressive strength values of concretes containing metakaolin and silica fume. The training and testing results in the neural network models have shown that neural networks have strong potential for predicting 1, 3, 7, 28, 56, 90 and 180 days compressive strength values of concretes containing metakaolin and silica fume.     

基于导波理论的管道缺陷成像研究

在导波传播理论和频散波形预测理论的基础上，研究了超声导波在检测管道时，如何利用时域波形重构缺陷图像的问题。由于根据导波检测理论可以获得在任意时间管道表面任意位置的波形信息，而且管道上缺陷的存在直接影响导波信号的幅值，并被包含在回波信号中为传感器接收，所以通过对接收信号进行相应的处理便可预测缺陷处的波形用以成像。针对管道中的单通孔人工缺陷和轴向位置不同的双槽形人工缺陷，利用非轴对称端面加载的传感器激励L（0，2）模态导波进行检测，通过改变传感器周向位置获得一组检测信号以实现缺陷形状的成像。成像结果不仅能反映缺陷的类型，还能实现缺陷在轴向和周向的二维定位，对轴向双缺陷的定位取决于时间参数的选择。     

Neural networks for load torque monitoring of an induction motor

In this paper, the possibility to use neural networks for the monitoring of the load torque of induction motors is investigated. In particular, unsupervised neural networks are used to detect possible torque anomalies and supervised neural networks are used to identify the average value of steady-state load torque. These networks are trained and validated on the data gathered from a 1.5 kW three-phase squirrel-cage induction motor. Their generalisation abilities have been tested through the data collected with a 3 kW induction motor.     

Novel heat dissipation design for light emitting diode applications

The authors offer a new design in support of efficient heat dissipation for light emitting diodes (LEDs). In the first part of this paper we discuss improvements in LED packaging materials and layer assembly, and then describe the addition of a thin layer of electroplated copper to the LED base assembly to reduce thermal resistance and increase thermal diffusion efficiency. Also described is a three-dimensional finite element simulation that we performed to verify the proposed design (0.75, 1, and 3 W LED chip temperatures) and a LED heat transfer behavior analysis. The results indicate that the addition of a 9 mm² electroplated copper layer to the LED base assembly improved LED thermal dissipation by reducing chip temperature by 5°C compared to LEDs without the copper layer packaging. In the second part of this paper we describe (a) our heat pipe system/heat sink design for LED illumination, and (b) experiments in which we changed both working fluid mass and rotation angle to determine their effects on heat pipe cooling. Our results indicate that the most efficient heat dissipation occurred when an added heat pipe was arranged horizontally. Good heat dissipation was observed for heat pipes containing 2.52 g of water (temperature reduced by 50°C). Larger water volumes failed to dissipate additional heat due to the presence of steam inside the pipe.     

Using artificial neural networks for classification of black-and-white images

This work is devoted to the problem of automatic classification of binary images. To solve this problem, neural networks are used. Experiments based on a database of faxgrams are performed with neural networks of different types. __________ Translated from Kibernetika i Sistemnyi Analiz, No. 2, pp. 184–187, March–April 2008.     

Database Integration Using Neural Networks: Implementation and Experiences

Applications in a wide variety of industries require access to multiple heterogeneous distributed databases. One step in heterogeneous database integration is semantic integration: identifying corresponding attributes in different databases that represent the same real world concept. The rules of semantic integration can not be ‘pre-programmed’ since the information to be accessed is heterogeneous and attribute correspondences could be fuzzy. Manually comparing all possible pairs of attributes is an unreasonably large task. We have applied artificial neural networks (ANNs) to this problem. Metadata describing attributes is automatically extracted from a database to represent their ‘signatures’. The metadata is used to train neural networks to find similar patterns of metadata describing corresponding attributes from other databases. In our system, the rules to determine corresponding attributes are discovered through machine learning. This paper describes how we applied neural network techniques in a database integration problem and how we represent an attribute with its metadata as discriminators. This paper focuses on our experiments on effectiveness of neural networks and each discriminator. We also discuss difficulties of using neural networks for this problem and our wish list for the Machine Learning community. Received 18 February 1999 / Revised 22 April 1999 / Accepted in revised form 20 November 1999     

Optimization of type-2 fuzzy weights in backpropagation learning for neural networks using GAs and PSO

In this paper the optimization of type-2 fuzzy inference systems using genetic algorithms (GAs) and particle swarm optimization (PSO) is presented. The optimized type-2 fuzzy inference systems are used to estimate the type-2 fuzzy weights of backpropagation neural networks. Simulation results and a comparative study among neural networks with type-2 fuzzy weights without optimization of the type-2 fuzzy inference systems, neural networks with optimized type-2 fuzzy weights using genetic algorithms, and neural networks with optimized type-2 fuzzy weights using particle swarm optimization are presented to illustrate the advantages of the bio-inspired methods. The comparative study is based on a benchmark case of prediction, which is the Mackey-Glass time series (for τ = 17) problem.     

基于神经网络的超声导波管道缺陷识别

在役石油管道的腐蚀是造成石油管线运输故障的重要原因,适时检测在役管道是否被腐蚀至关重要。研究了超声导波进行长距离在役管道检测技术,并利用人工神经网络进行管道缺陷的智能识别,通过超声导波设备进行了管道缺陷检测实验,从原始检测数据的信号处理结果中提取出了样本特征值,并建立和训练了一种用于实现管道缺陷识别的BP神经网络。实验表明:使用该网络可进行超声导波管道缺陷的自动识别。     

A Study of Hybrid Neural Network Approaches and the Effects of Missing Data on Traffic Forecasting

In this paper we present an application of hybrid neural network approaches and an assessment of the effects of missing data on motorway traffic flow forecasting. Two hybrid approaches are developed using a Self-Organising Map (SOM) to initially classify traffic into different states. The first hybrid approach includes four Auto-Regressive Integrated Moving Average (ARIMA) models, whilst the second uses two Multi-Layer Perception (MLP) models. It was found that the SOM/ARIMA hybrid approach out-performs all individual ARIMA models, whilst the SOM/MLP hybrid approach achieves superior forecasting performance to all models used in this study, including three na?ve models. The effects of different proportions of missing data on Neural Network (NN) performance when forecasting traffic flow are assessed and several initial substitution options to replace missing data are discussed. Over-all, it is shown that ARIMA models are more sensitive to the percentage of missing data than neural networks in this context.       

Model study for rehabilitation planning of water supply network

Inefficient delivery and inadequate coverage of water supply and sanitation services are major concerns for public health in the urban regions of developing countries. The contamination of the treated water within distribution system leads to frequent outbreaks of waterborne diseases. This demands a proactive approach that is based on risk assessment and a management methodology to ensure water safety. In this context, the paper presents a model study undertaken for the water supply network of a pilot area in Hyderabad city, India. The risk assessment methodology uses geospatial databases of the water supply network, sewer network, open drains, groundwater table, pressure in pipes, and soil data with a number of system-specific attributes. Fuzzy multi-criteria evaluation approach with qualitative and quantitative domain knowledge is employed in pipe condition assessment model. The physical parameters viz. pipe age, material, diameter; operational parameters viz. intermittency, number of breaks and bursts, and leakage in the system; and environmental parameters viz. workmanship, bedding condition, and traffic determine the vulnerability of pipes to contaminant intrusion. The zones of contamination formed in the soil near open drains and sewer crossings (i.e., hazards) are delineated using a contaminant ingress model. The risk of contaminant intrusion is assessed as a function of vulnerability and hazard. The results indicate that roughly 3% of pipes in the network are in Bad condition and require rehabilitation on a priority basis; about 46% of pipes are in Medium condition. The study describes a techno-economically feasible approach to assist water managers and policy makers in delivering safe drinking water.     

A deep learning-based PM2.5 concentration estimator

PM2.5 does great harm to human beings. In particular, it can lead to an increase in human lung cancer. In this paper, we propose a PM2.5 concentration estimator based on deep convolutional neural networks. The proposed method consists of three modules. First, we generate a hallucinated reference image of PM2.5 by using deep convolutional neural networks. The discrepancy map of the PM2.5 image and the hallucinated reference image are calculated. Second, the discrepancy map and the distorted PM2.5 image are used to extract the features. Third, the prediction module based on neural networks utilizes those extracted features to predict PM2.5 concentrations. Compared to existing PM2.5 estimators and state-of-art image quality assessment(IQA) metrics, experimental results illustrate the effectiveness of the proposed model on the AQID database.     

Prediction of artificial soil’s unconfined compression strength test using statistical analyses and artificial neural networks

Laboratory prediction of the unconfined compression strength (UCS) of cohesive soils is important to determine the shear strength properties. However, this study presents the application of different methods simple–multiple analysis and artificial neural networks for the prediction of the UCS from basic soil properties. Regression analysis and artificial neural networks prediction indicated that there exist acceptable correlations between soil properties and unconfined compression strength. Besides, artificial neural networks showed a higher performance than traditional statistical models for predicting UCS. Regression analysis and artificial neural network prediction indicated strong correlations (R² = 0.71–0.97) between basic soil properties and UCS. It has been shown that the correlation equations obtained by regression analyses are found to be reliable in practical situations.     

Dealing with missing values in neural network-based diagnostic systems

Backpropagation neural networks have been applied to prediction and classification problems in many real world situations. However, a drawback of this type of neural network is that it requires a full set of input data, and real world data is seldom complete. We have investigated two ways of dealing with incomplete data — network reduction using multiple neural network classifiers, and value substitution using estimated values from predictor networks — and compared their performance with an induction method. On a thyroid disease database collected in a clinical situation, we found that the network reduction method was superior. We conclude that network reduction can be a useful method for dealing with missing values in diagnostic systems based on backpropagation neural networks.     

基于GPS技术的地埋管道杂散电流检测设备

在电气化轨道交通沿线,有列车经过的时候,轨道电流必将在地埋管道上产生较大的杂散电流,从而使地埋管道产生电化学腐蚀,对管道产生极大的破坏。因此,为找出腐蚀点的具体位置,设计了本检测设备,包括中断器及管道电位检测仪。设备以高速微处理器为系统核心,应用GPS技术,通过同步通断测量技术准确测量管道电位,并选用液晶屏及薄膜按键进行了良好的人机交互设计。     

Deep-SBFL:基于频谱的深度神经网络缺陷定位方法

深度神经网络已经在自动驾驶和智能医疗等领域取得了广泛的应用.与传统软件一样,深度神经网络也不可避免地包含缺陷,如果做出错误决定,可能会造成严重后果.因此,深度神经网络的质量保障受到了广泛关注.然而,深度神经网络与传统软件存在较大差异,传统软件质量保障方法无法直接应用于深度神经网络,需要设计有针对性的质量保障方法.软件缺陷定位是保障软件质量的重要方法之一,基于频谱的缺陷定位方法在传统软件的缺陷定位中取得了很好的效果,但无法直接应用于深度神经网络.在传统软件缺陷定位方法的基础上提出了一种基于频谱的深度神经网络缺陷定位方法 Deep-SBFL.该方法首先通过收集深度神经网络的神经元输出信息和预测结果作为频谱信息;然后将频谱信息进行处理作为贡献信息,以用于量化神经元对预测结果所做的贡献;最后提出了针对深度神经网络缺陷定位的怀疑度公式,基于贡献信息计算深度神经网络中神经元的怀疑度并进行排序,以找出最有可能存在缺陷的神经元.为验证该方法的有效性,以E_Inspect@n (结果排序列表前n个位置内成功定位的缺陷数)和EXAM (在找到缺陷元素之前必须检查元素的百分比)作为评测指...     

Polynomial-based radial basis function neural networks (P-RBF NNs) and their application to pattern classification

Polynomial neural networks have been known to exhibit useful properties as classifiers and universal approximators. In this study, we introduce a concept of polynomial-based radial basis function neural networks (P-RBF NNs), present a design methodology and show the use of the networks in classification problems. From the conceptual standpoint, the classifiers of this form can be expressed as a collection of “if-then” rules. The proposed architecture uses two essential development mechanisms. Fuzzy clustering (Fuzzy C-Means, FCM) is aimed at the development of condition parts of the rules while the corresponding conclusions of the rules are formed by some polynomials. A detailed learning algorithm for the P-RBF NNs is developed. The proposed classifier is applied to two-class pattern classification problems. The performance of this classifier is contrasted with the results produced by the “standard” RBF neural networks. In addition, the experimental application covers a comparative analysis including several previous commonly encountered methods such as standard neural networks, SVM, SOM, PCA, LDA, C4.5, and decision trees. The experimental results reveal that the proposed approach comes with a simpler structure of the classifier and better prediction capabilities.