Paper manufacturing is energy demanding and improvedmodelling of the pulp bleach process is the main non-invasive means ofreducing energy costs. In this paper, time it takes to bleach paper pulpto desired brightness is examined. The model currently used is analysedand benchmarked against two machine learning models (Random Forestand TreeBoost). Results suggests that the current model can be super-seded by the machine learning models and it does not use the optimalcompact subset of features. Despite the differences between the machinelearning models, a feature ranking correlation has been observed for thenew models. One novel, yet unused, feature that both machine learningmodels found to be important is the concentration of bleach agent.
In a global economy, manufacturers mainly compete with cost efficiency of production, as the price of raw materials are similar worldwide. Heavy industry has two big issues to deal with. On the one hand there is lots of data which needs to be analyzed in an effective manner, and on the other hand making big improvements via investments in cooperate structure or new machinery is neither economically nor physically viable. Machine learning offers a promising way for manufacturers to address both these problems as they are in an excellent position to employ learning techniques with their massive resource of historical production data. However, choosing modelling a strategy in this setting is far from trivial and this is the objective of this article. The article investigates characteristics of the most popular classifiers used in industry today. Support Vector Machines, Multilayer Perceptron, Decision Trees, Random Forests, and the meta-algorithms Bagging and Boosting are mainly investigated in this work. Lessons from real-world implementations of these learners are also provided together with future directions when different learners are expected to perform well. The importance of feature selection and relevant selection methods in an industrial setting are further investigated. Performance metrics have also been discussed for the sake of completion.
Radar speed warning signs (RSWS) have been used in recent years across Sweden and elsewhere in the world. Such signs measure vehicle speed using radar and are designed to display a message when the driver exceeds a pre-set threshold speed, which is often relative to the speed limit on a particular road segment. RSWS are typically placed on locations which are perceived to be problematic by relevant authorities. Excessive speeding or road accidents are examples of such perceived problems. Deploying RSWS in many relevant locations is often impractical due to the lack of necessary power supply needed for operation. Battery driven RSWS are an alternative but are less attractive because of limited running time and frequent maintenance (changing batteries etc). Therefore, solar powered RSWS are more desirable. However, these signs are also dependent on batteries that need to be charged. The duration of operation of solar powered RSWS largely depend on how often the sign is triggered. Constant activation of the sign drains the battery. It is desirable to trigger the sign only when necessary. Hence, the main goal of this research is to design a model that optimises the performance of RSWS depending on prevailing conditions i.e traffic flows during different times of the day and so on. Vehicle speed data had been collected at a test site in Sweden all hours of the day. This paper attempts to use a hybrid system based on Apriori and K-means clustering algorithm. Apriori algorithm is simple and efficient to determine associations’ rules among attributes in particular to discover the most common combination that can occur within the data set. K-means clustering is basically used to quantize the input variables into smaller clusters that can easily derive the trigger threshold value. The proposed hybrid system indicated that the system was able to trigger solar RWWS efficiently.
Excessive or inappropriate speeds are a key factor in traffic fatalities and crashes. Vehicle-activated signs (VASs) are therefore being extensively used to reduce speeding to increase traffic safety. A VAS is triggered by an individual vehicle when the driver exceeds a speed threshold, otherwise known as trigger speed (TS). The TS is usually set to a constant, normally proportional to the speed limit on the particular segment of road. Decisions concerning the TS largely depend on the local traffic authorities. The primary objective of this article is to help authorities determine the TS that gives an optimal effect on the Mean and Standard Deviation of speed. The data were systematically collected using radar technology whilst varying the TS. The results show that when the applied TS was set near the speed limit, the standard deviation was high. However, the Standard Deviation decreased substantially when the threshold was set to the 85th percentile. This decrease occurred without a significant increase in the mean speed. It is concluded that the optimal threshold speed should approximate the 85th percentile, though VASs should ideally be individually calibrated to the traffic conditions at each site.
Optimal trigger speeds for vehicle activated signs were not considered in previous studies. The main aim of this paper is to summarise the findings of optimum trigger speed for vehicle activated signs. A secondary aim is to be able to build and report a dynamic trigger speed based on an accurate predictive model to be able to trigger operation of vehicle activated signs. A data based calibration method for the radar used in the experiment has been developed and evaluated. Results from the study indicate that the optimal trigger speed should be primarily aimed at lowering the standard deviation. Results also indicate that the optimal trigger speed should be set near the 85th percentile speed, to be able to lower the standard deviation. A comparative study investigating the use of several predictive models showed that random forest is an appropriate model to dynamically predict trigger speeds.
Vehicle-activated signs (VAS) are speed-warning signs activated by radar when the driver speed exceeds a pre-set threshold, i.e. the trigger speed. The trigger speed is often set relative to the speed limit and is displayed for all types of vehicles. It is our opinion that having a static setting for the trigger speed may be inappropriate, given that traffic and road conditions are dynamic in nature. Further, different vehicle classes (mainly cars and trucks) behave differently, so a uniform trigger speed of such signs may be inappropriate to warn different types of vehicles. The current study aims to investigate an automatic VAS, i.e. one that could warn vehicle users with an appropriate trigger speed by taking into account vehicle types and road conditions. We therefore investigated different vehicle classes, their speeds, and the time of day to be able to conclude whether different trigger speeds of VAS are essential or not. The current study is entirely data driven; data are initially presented to a self-organising map (SOM) to be able to partition the data into different clusters, i.e. vehicle classes. Speed, time of day, and length of vehicle were supplied as inputs to the SOM. Further, the 85th percentile speed for the next hour is predicted using appropriate prediction models. Adaptive neuro-fuzzy inference systems and random forest (RF) were chosen for speed prediction; the mean speed, traffic flow, and standard deviation of vehicle speeds were supplied as inputs for the prediction models. The results achieved in this work show that RF is a reliable model in terms of accuracy and efficiency, and can be used in finding appropriate trigger speeds for an automatic VAS.
Vehicle activated signs and Speed indicator devices are safety signs used to warn and remind drivers that they are exceeding the speed limit on a particular road segment. This article has analysed and compared such signs with the aim of reporting the most suitable sign for relevant situations. Vehicle speeds were recorded at different test sites and the effects of the signs were studied by analyzing the mean and standard deviation. Preliminary results from the work indicate that both types of signs have variable effects on the mean and standard deviation of speed on a given road segment. Speed indicator devices were relatively more effective than vehicle activated signs on local roads; in contrast their effectivity was only comparable when tested on highways.
This paper reviews the effectiveness of vehicle activated signs. Vehicle activated signs are being reportedly used in recent years to display dynamic information to road users on an individual basis in order to give a warning or inform about a specific event. Vehicle activated signs are triggered individually by vehicles when a certain criteria is met. An example of such criteria is to trigger a speed limit sign when the driver exceeds a pre-set threshold speed. The preset threshold is usually set to a constant value which is often equal, or relative, to the speed limit on a particular road segment.
This review examines in detail the basis for the configuration of the existing sign types in previous studies and explores the relation between the configuration of the sign and their impact on driver behavior and sign efficiency. Most of previous studies showed that these signs have significant impact on driver behavior, traffic safety and traffic efficiency. In most cases the signs deployed have yielded reductions in mean speeds, in speed variation and in longer headways. However most experiments reported within the area were performed with the signs set to a certain static configuration within applicable conditions. Since some of the aforementioned factors are dynamic in nature, it is felt that the configurations of these signs were thus not carefully considered by previous researchers and there is no clear statement in the previous studies describing the relationship between the trigger value and its consequences under different conditions. Bearing in mind that different designs of vehicle activated signs can give a different impact under certain conditions of road, traffic and weather conditions the current work suggests that variable speed thresholds should be considered instead.
Accurate speed prediction is a crucial step in the development of a dynamic vehcile activated sign (VAS). A previous study showed that the optimal trigger speed of such signs will need to be pre-determined according to the nature of the site and to the traffic conditions. The objective of this paper is to find an accurate predictive model based on historical traffic speed data to derive the optimal trigger speed for such signs. Adaptive neuro fuzzy (ANFIS), classification and regression tree (CART) and random forest (RF) were developed to predict one step ahead speed during all times of the day. The developed models were evaluated and compared to the results obtained from artificial neural network (ANN), multiple linear regression (MLR) and naïve prediction using traffic speed data collected at four sites located in Sweden. The data were aggregated into two periods, a short term period (5-min) and a long term period (1-hour). The results of this study showed that using RF is a promising method for predicting mean speed in the two proposed periods.. It is concluded that in terms of performance and computational complexity, a simplistic input features to the predicitive model gave a marked increase in the response time of the model whilse still delivering a low prediction error.
Solar-powered vehicle activated signs (VAS) are speed warning signs powered by batteries that are recharged by solar panels. These signs are more desirable than other active warning signs due to the low cost of installation and the minimal maintenance requirements. However, one problem that can affect a solar-powered VAS is the limited power capacity available to keep the sign operational. In order to be able to operate the sign more efficiently, it is proposed that the sign be appropriately triggered by taking into account the prevalent conditions. Triggering the sign depends on many factors such as the prevailing speed limit, road geometry, traffic behaviour, the weather and the number of hours of daylight. The main goal of this paper is therefore to develop an intelligent algorithm that would help optimize the trigger point to achieve the best compromise between speed reduction and power consumption. Data have been systematically collected whereby vehicle speed data were gathered whilst varying the value of the trigger speed threshold. A two stage algorithm is then utilized to extract the trigger speed value. Initially the algorithm employs a Self-Organising Map (SOM), to effectively visualize and explore the properties of the data that is then clustered in the second stage using K-means clustering method. Preliminary results achieved in the study indicate that using a SOM in conjunction with K-means method is found to perform well as opposed to direct clustering of the data by K-means alone. Using a SOM in the current case helped the algorithm determine the number of clusters in the data set, which is a frequent problem in data clustering.
The accurate measurement of a vehicle’s velocity is an essential feature in adaptive vehicle activated sign systems. Since the velocities of the vehicles are acquired from a continuous wave Doppler radar, the data collection becomes challenging. Data accuracy is sensitive to the calibration of the radar on the road. However, clear methodologies for in-field calibration have not been carefully established. The signs are often installed by subjective judgment which results in measurement errors. This paper develops a calibration method based on mining the data collected and matching individual vehicles travelling between two radars. The data was cleaned and prepared in two ways: cleaning and reconstructing. The results showed that the proposed correction factor derived from the cleaned data corresponded well with the experimental factor done on site. In addition, this proposed factor showed superior performance to the one derived from the reconstructed data.
Mathematical analytical modeling and computer simulation of the physiological system is a complex problem with great number of variables and equations. The objective of this research is to describe the insulin-glucose subsystem using multi-agent modeling based on intelligence agents. Such an approach makes the modeling process easier and clearer to understand; moreover, new agents can be added or removed more easily to any investigations. The Stolwijk-Hardy mathematical model is used in two ways firstly by simulating the analytical model and secondly by dividing up the same model into several agents in a multiagent system. In the proposed approach a multi-agent system was used to build a model for glycemic homeostasis. Agents were used to represent the selected elements of the human body that play an active part in this process. The experiments conducted show that the multi-agent model has good temporal stability with the implemented behaviors, and good reproducibility and stability of the results. It has also shown that no matter what the order of communication between agents, the value of the result of the simulation was not affected. The results obtained from using the framework of multi-agent system actions were consistent with the results obtained with insulin-glucose models using analytical modeling.
National Railway Administrations in Northern Europe do not employ systematic procedures in monitoring the current state of vegetation to form the basis of maintenance decision making. Current day vegetation maintenance is largely based on human visual estimates. This paper investigates a machine vision (MV) approach to be able to automatically quantify the amount of vegetation on a given railway section. An investigation assessing the reliability of human estimates is also conducted along the same railway section.A machine vision algorithm was developed and implemented. Initially, the algorithm determines a region of interest (ROI), i.e. the desired monitored area in each collected image. This ROI is dependent on fixed objects in the image, namely the two rails. When the rails are found the algorithm will compute the ROI, which is predetermined by e.g. the railway administrator. After this, a perspective projection correction will be made, and the vegetation will be segmented. Cover is reported as a percentage of the total ROI for each image. Results: The machine vision algorithm is capable of processing 98% of the images. Failure in the remaining 2% of cases is attributed to the algorithms' inability in find the rails within the image. Analysis of variance tests were conducted to compare the observers cover assessments in sample plots. Upon comparing the observers plot wise mean estimates with the machine vision output, results show that the human visual estimates do not correlate with the results reported by the machine vision output. As such, the result indicates that it is very hard to fit human estimates by regression with the machine vision result. Additionally the results show that humans are not in agreement with each other, and often are exaggerating the extent of vegetation cover compared to the machine vision output.The investigation shows that one should be very careful when trusting/interpreting human visual estimates. In conclusion, based on the results, the automated machine vision solution is proposed as complementing, or replacing, manual human inspections serving as a base for vegetation control decisions. Impact: By objectively measuring the quantity of vegetation, the maintenance planning and procurement can be effectively improved over time. A machine vision approach for condition monitoring of vegetation will enable condition based maintenance with prior consideration on issues mainly relevant to vegetation type, quantity and biodiversity.
The national railway administrations in Scandinavia, Germany, and Austria mainly resort to manual inspections to control vegetation growth along railway embankments. Manually inspecting railways is slow and time consuming. A more worrying aspect concerns the fact that human observers are often unable to estimate the true cover of vegetation on railway embankments. Further human observers often tend to disagree with each other when more than one observer is engaged for inspection. Lack of proper techniques to identify the true cover of vegetation even result in the excess usage of herbicides; seriously harming the environment and threating the ecology. Hence work in this study has investigated aspects relevant to human variationand agreement to be able to report better inspection routines. This was studied by mainly carrying out two separate yet relevant investigations.First, thirteen observers were separately asked to estimate the vegetation cover in nine imagesacquired (in nadir view) over the railway tracks. All such estimates were compared relatively and an analysis of variance resulted in a significant difference on the observers’ cover estimates (p<0.05). Bearing in difference between the observers, a second follow-up field-study on the railway tracks was initiated and properly investigated. Two railway segments (strata) representingdifferent levels of vegetationwere carefully selected. Five sample plots (each covering an area of one-by-one meter) were randomizedfrom each stratumalong the rails from the aforementioned segments and ten images were acquired in nadir view. Further three observers (with knowledge in the railway maintenance domain) were separately asked to estimate the plant cover by visually examining theplots. Again an analysis of variance resulted in a significant difference on the observers’ cover estimates (p<0.05) confirming the result from the first investigation.The differences in observations are compared against a computer vision algorithm which detects the "true" cover of vegetation in a given image. The true cover is defined as the amount of greenish pixels in each image as detected by the computer vision algorithm. Results achieved through comparison strongly indicate that inconsistency is prevalent among the estimates reported by the observers. Hence, an automated approach reporting the use of computer vision is suggested, thus transferring the manual inspections into objective monitored inspections
This paper investigates problems concerning vegetation along railways and proposes automatic means of detecting ground vegetation. Digital images of railway embankments have been acquired and used for the purpose. The current work mainly proposes two algorithms to be able to achieve automation. Initially a vegetation detection algorithm has been investigated for the purpose of detecting vegetation. Further a rail detection algorithm that is capable of identifying the rails and eventually the valid sampling area has been investigated. Results achieved in the current work report satisfactory (qualitative) detection rates.
Vegetation growing on railway trackbeds and embankments can present several potential problems. Consequently, such vegetation iscontrolled through various maintenance procedures. In order to investigate the extent of maintenance needed, one of the first steps in anymaintenance procedure is to monitor or inspect the railway section in question. Monitoring is often carried out manually by sending out inspectorsor by watching recorded video clips of the section in question.To facilitate maintenance planning, the ability to assess the extent of vegetation becomes important. This paper investigates the reliability ofhuman assessments of vegetation on railway trackbeds.In this study, five maintenance engineers made independent visual estimates of vegetation cover and counted the number of plant clusters fromimages.The test results showed an inconsistency between the raters when it came to visually estimating plant cover and counting plant clusters. The resultsshowed that caution should be exercised when interpreting individual raters’ assessments of vegetation.
Increasing costs and competitive business strategies are pushing sawmill enterprises to make an effort for optimization of their process management. Organizational decisions mainly concentrate on performance and reduction of operational costs in order to maintain profit margins. Although many efforts have been made, effective utilization of resources, optimal planning and maximum productivity in sawmill are still challenging to sawmill industries. Many researchers proposed the simulation models in combination with optimization techniques to address problems of integrated logistics optimization. The combination of simulation and optimization technique identifies the optimal strategy by simulating all complex behaviours of the system under consideration including objectives and constraints. During the past decade, an enormous number of studies were conducted to simulate operational inefficiencies in order to find optimal solutions. This paper gives a review on recent developments and challenges associated with simulation and optimization techniques. It was believed that the review would provide a perfect ground to the authors in pursuing further work in optimizing sawmill yard operations.
This paper reports the findings of using multi-agent based simulation model to evaluate the sawmill yard operations within a large privately owned sawmill in Sweden, Bergkvist Insjön AB in the current case. Conventional working routines within sawmill yard threaten the overall efficiency and thereby limit the profit margin of sawmill. Deploying dynamic work routines within the sawmill yard is not readily feasible in real time, so discrete event simulation model has been investigated to be able to report optimal work order depending on the situations. Preliminary investigations indicate that the results achieved by simulation model are promising. It is expected that the results achieved in the current case will support Bergkvist-Insjön AB in making optimal decisions by deploying efficient work order in sawmill yard.
This paper summarises the results of using image processing technique to get information about the load of timber trucks before their arrival using digital images or geo tagged images. Once the images are captured and sent to sawmill by drivers from forest, we can predict their arrival time using geo tagged coordinates, count the number of (timber) logs piled up in a truck, identify their type and calculate their diameter. With this information we can schedule and prioritise the inflow and unloading of trucks in the light of production schedules and raw material stocks available at the sawmill yard. It is important to keep all the actors in a supply chain integrated coordinated, so that optimal working routines can be reached in the sawmill yard.
Bin planning (arrangements) is a key factor in the timber industry. Improper planning of the storage bins may lead to inefficient transportation of resources, which threaten the overall efficiency and thereby limit the profit margins of sawmills. To address this challenge, a simulation model has been developed. However, as numerous alternatives are available for arranging bins, simulating all possibilities will take an enormous amount of time and it is computationally infeasible. A discrete-event simulation model incorporating meta-heuristic algorithms has therefore been investigated in this study. Preliminary investigations indicate that the results achieved by GA based simulation model are promising and better than the other meta-heuristic algorithm. Further, a sensitivity analysis has been done on the GA based optimal arrangement which contributes to gaining insights and knowledge about the real system that ultimately leads to improved and enhanced efficiency in sawmill yards. It is expected that the results achieved in the work will support timber industries in making optimal decisions with respect to arrangement of storage bins in a sawmill yard.
This paper addresses and deals with the problem of automating condition monitoring of wood in the transportation domain. Current day condition monitoring applications involving wood are mostly carried out through visual inspection and if necessary some impact acoustic examination is carried out. These inspections are mostly done intuitively by skilled personnel. Hence, it is desired to automate such intuitive human skills for the development of more robust and reliable testing methods. Data resulting from impact acoustics tests made on wooden beams has been used. The relation between condition of the wooden beam and their respective emissions has been analyzed experimentally applying different feature extraction techniques. Combining the usage of traditional frequency extraction techniques like the magnitude of the signal together with famous speech recognition techniques like Cepstral Coefficients, Linear Predictive Coding yield good results. Effect of using classifiers like Gaussian Mixture Models and Learning Vector Quantization has been tested and compared. In the current case Gaussian mixture model seem to achieve higher classification rates than Learning Vector Quantization model.
The thesis aims to investigate the current state of railway sleeper inspection and proposes automatic testing procedures based on pattern recognition for future inspections concerning the condition of the sleeper. Wooden railway sleeper inspections in Sweden are currently done by hand. That is to say, a human inspector in charge of the maintenance activities visually examines each structure in turn for the presence of cracks on the sleeper. Where necessary some deeper inspection may be carried out on site, for example using an axe to hit and judge the condition of the sleeper by listening to the sound produced. Though the manual procedure uses non-destructive testing methods (visual and sound analysis), decision making is largely based on intuition; moreover the process is rather slow, expensive and also requires skilled and trained staff. Maintaining an even quality standard is another serious issue. In order to be able to fulfil the aims of the thesis, emphasis on the likely factors concerning sleeper condition was a key issue. Studies based on emulation of the human inspection process have been considered a promising route of enquiry. The emulation process is achieved by selecting and evaluating two non-destructive testing methods. The first method (impact acoustic analysis) aims to build an automatic procedure to replace the usage of an axe for distinguishing sounds; which can be described qualitatively as a crisp sound in case of a good sleeper and a dull thud on their bad counterparts. The second method (vision analysis) is to develop an appropriate machine vision algorithm to replicate the visual examination. Data were collected for each of the above methods and appropriate features were extracted. Frequency based features and crack based features have been extracted in the case of impact acoustics and machine vision methods respectively. Pattern recognition has been mainly researched for further classification work concerning the condition (good or bad) of the sleeper. Research conducted on the usage of the inspection methods such as impact acoustic and machine vision analysis show that the methods can form the basis of an automatic sleeper condition monitoring procedure. Further, two more non-destructive testing methods namely electrical resistivity analysis and ultrasound analysis have also been tested. Usage of such methods did not yield success in the current case, but they have contributed in adding knowledge to the domain in cases of relevant problems. Initially, work has pursued data from only one inspection method at a time. Given that data from a single method (or sensor) seems not to be adequate to make a reliable judgement; data fusion was investigated with an aim of achieving more reliable and robust results. Data fusion has been investigated at three different levels namely sensor-level fusion, feature-level fusion and classifier-level fusion. Results achieved by fusion in the current thesis demonstrate an overall efficiency of around 90% when compared to a human operator. This can be regarded as a good result, given that even humans disagree on certain judgements; and destructive testing can be seen as the only way to resolve such disagreements.
Wooden railway sleeper inspections in Sweden and to a large extent elsewhere are carried out manually by a human operator; visual inspection being the most common approach. Manually inspecting railway sleepers is slow and time consuming. Machine vision algorithms investigating surface cracks on the sleeper and sinking of the metal plate have been studied for the purpose of automating the task. In this particular article, information concerning how far the fastening nail has lifted out of position is investigated with an aim of using such information while assessing the condition of the sleeper. Laser beams incident on the sleeper have been used to highlight the geometrical form of the sleeper/plate/nail complex. Digital images of the nail were acquired mimic human visual capabilities. Appropriate image analysis techniques were applied to further process the images and necessary features were extracted. Results of unsupervised learning, achieved in the current work indicate that expectation maximization algorithm produced reliable results.
This article describes a method of automatically detecting, counting and classifying logs on a timber truck using a photograph (taken by the driver). An image-processing algorithm is developed to process the photograph to calculate an estimate of the number of logs present and their respective diameters. The algorithm uses color information in multiple color spaces as well as geometrical operators to segment the image and extract the relevant information. This information enables the sawmill to better plan internal logistics and production in advance of the truck’s arrival time. The algorithm is robust with respect to external factors such as varying lighting conditions and camera angle, but some inaccuracies remain, mainly caused by logs being occluded or covered in mud or snow.
This paper attempts to summarise the findings of a large number of research papers deploying artificial intelligence (AI) techniques for the automatic interpretation of data from non-destructive testing (NDT). Problems in the rail transport domain are mainly discussed. However, a majority of the emphasis in this paper is laid on rail inspection problems, since it was believed that the review would provide a perfect ground to the authors in pursuing further work within the rail inspection area. NDT is a broad name for a variety of methods and procedures concerned with all aspects of uniformity, quality and serviceability of materials and structures, without causing damage to the material that is being inspected. During the past several years, problems concerning the automatic interpretation of data from NDT have received good attention and have stimulated interests in other areas like transportation, for making key assessments within some of its subject areas. Rail, air and marine industries together with bridge inspection and pavement maintenance are good examples of such areas where a considerable amount of work has been done. Such work neatly splits into two schools. The first school investigates the classical usage of data by an experienced human operator to determine the condition of the inspected material or structure. The other school focuses attention on the automatic interpretation of NDT data using AI techniques, in determining the result of inspection. The scope of this paper is only limited to the automatic interpretation of data from NDT, with the goal of assessing embedded flaws as quickly and accurately as possible in a cost effective fashion. AI techniques such as neural networks, machine vision, knowledge-based systems and fuzzy logic were applied to a wide spectrum of problems in the area. A secondary goal was to provide an insight into possible research methods concerning railway sleeper inspection by automatic interpretation of data. A brief introduction is provided for the benefit of the readers unfamiliar with the techniques.
Wooden railway sleeper inspections in Sweden are currently performed manually by a human operator; such inspections are to large extent based on visual analysis. In this paper a machine vision based approach has been considered to emulate the visual abilities of the human operator to enable automation of the process. Digital images from either ends (left and right) of the sleepers have been acquired. A pattern recognition approach has been adopted to classify the condition of the sleeper into classes (good or bad) and thereby achieve automation. Appropriate image analysis techniques were applied and relevant features such as the number of cracks on a sleeper, average length and width of the crack and the condition of the metal plate were determined. Feature fusion has been proposed in order to integrate the features obtained from each end for the classification task which follows. The effect of using classifiers like multi-layer perceptron and support vector machines has been tested and compared. Results obtained from the experiments show that multi-layer perceptron and support vector machines have achieved encouraging results, with a classification accuracy of 90%; thereby exhibiting a competitive performance when compared to a human operator.
Wooden railway sleeper inspections in Sweden are currently done by hand. That is to say, a human inspector in charge of the maintenance activities visually examines each structure in turn for the presence of cracks on the sleeper. Where necessary some deeper inspection may be carried out on site, for example using an axe to hit and judge the condition of the sleeper by listening to the sound produced. Though the manual procedure uses non-destructive testing methods (visual and sound analysis), decision-making is largely based on intuition; moreover the process is rather slow, expensive and also requires skilled and trained staff. Maintaining an even quality standard is another serious issue. Hence, it is desired to automate the human inspection process by proposing automatic testing procedures for future inspections concerning the condition of the sleeper. Studies based on emulation of the human inspection process have been considered a promising route of enquiry for automation. Such an emulation process is achieved by selecting and evaluating two non-destructive inspection methods. The first method (impact acoustic analysis) aims to build an automatic system to replace the usage of an axe for distinguishing sounds. The second method (visual analysis) is to develop an appropriate machine vision algorithm to replicate the visual examination. Further, the above-mentioned methods were fused (data fusion) to generate a single output condition concerning the condition of the sleeper. In the current work, fusion has been achieved in mainly three levels, namely sensor-level, feature-level and classifier-level. Experimental results achieved in this work indicate that data fusion has achieved superior performance when compared with using data from one method at a time.
Railway sleepers are a key engineering element of all railways. Lack of much sophistication in monitoring railway sleepers makes it a key problem within the rail transportation domain. Current day condition monitoring applications involving wooden railway sleepers are mostly carried out through visual inspection and if necessary some impact acoustic examination is carried out. Decision making is largely based on intuition; moreover the process of manually inspecting sleepers is rather slow and expensive. Maintaining an even quality standard is another serious issue. In this article, a pattern recognition and classification approach is taken to automate such intuitive human skills for the development of more robust and reliable testing methods. Features were extracted from the impact acoustic emissions of wooden sleepers and were used for pattern classification. Time-frequency based feature extraction techniques such short-time Fourier transform and discrete wavelet transform yielded good results. Multi-layer perceptron, radial basis function neural networks and support vector machine classifiers have been tested and compared. Further classifier fusion was investigated by considering the output of single best classifiers as input to a new classifier with an aim of improving performance. Results obtained experimentally demonstrate a classification accuracy of around 84%.
Condition monitoring applications deploying the usage of impact acoustic techniques are mostly done intuitively by skilled personnel. In this article, a pattern recognition approach is taken to automate such intuitive human skills for the development of more robust and reliable testing methods. The focus of this work is to use the approach as a part of a major research project in the rail inspection area, within the domain of Intelligent Transport Systems. Data from impact acoustic tests made on wooden beams have been used. The relation between condition of the wooden beams and respective sounds they make when struck, has been analyzed experimentally. Features were extracted from the acoustic emissions of wooden beams and were used for pattern classification. Features such as magnitude of the signal, natural logarithm of the magnitude and Mel-frequency cepstral coefficients, yielded good results. The extracted feature vectors were used as input to various pattern classifiers for further pattern recognition task. The effect of using classifiers like Support vector machines and Multi-layer perceptron has been tested and compared. Results obtained experimentally, demonstrate that Support vector machines provide good detection rates for the classification of impact acoustic signals in the NDT domain.
Current day condition monitoring applications involving wood are mostly carried out through visual inspection and if necessary some impact acoustic examination is carried out. These inspections are mainly done intuitively by skilled personnel. In this paper, a pattern recognition approach has been considered to automate such intuitive human skills for the development of robust and reliable methods within the area. The study presents a comparison of several pattern recognition techniques combined with various stationary feature extraction techniques for classification of impact acoustic emissions. Further issues concerning feature fusion are discussed as well. It is hoped that this kind of broad analysis could be used to handle a wide spectrum of tasks within the area, and would provide a perfect ground for future research directions. A brief introduction to the techniques is provided for the benefit of the readers unfamiliar with the techniques. Pattern classifiers such as support vector machines, etc. are combined with stationary feature extraction techniques such as linear predictive cepstral coefficients, etc. Results from support vector machines in combination with linear predictive cepstral coefficients delivered good classification rates. However, Gaussian mixture models delivered higher classification rates when feature fusion is proposed.
This study has investigated the quality and reliability of manual assessments on railway embankments within the domain of railway maintenance. Manually inspecting vegetation on railway embankments is slow and time consuming. Maintenance personnel also require extensive knowledge of the plant species, ecology and bio-diversity to be able to recommend appropriate maintenance action. The overall objective of the study is to investigate the reliable nature of manual inspection routines in favour an automatic approach. Visual estimates of plant cover reported by domain experts’ have been studied on two separate railway sections in Sweden. The first study investigated visual estimates using aerial foliar cover (AFC) and sub-plot frequency (SF) methods to assess the plant cover on a railway section in Oxberg, Alvdalsbanan, Sweden. The second study investigated visual estimates using aerial canopy cover method on a railway section outside Vetlanda, Sweden. Visual estimates of the domain experts were recorded and analysis-of-variance (ANOVA) tests on the mean estimates were investigated to see whether if there were disagreements between the raters’. ICC(2, 1) was used to study the differences between the estimates. Results achieved in this work indicate statistically significant differences in the mean estimates of cover (p < 0.05) reported by the domain experts on both the occasions.
Current day vegetation assessments within railway maintenance are (to a large extent) carried out manually. This study has investigated the reliability of such manual assessments by taking three non-domain experts into account. Thirty-five track images under different conditions were acquired for the purpose. For each image, the raters’ were asked to estimate the cover of woody plants, herbs and grass separately (in %) using methods such as aerial canopy cover, aerial foliar cover and sub-plot frequency. Visual estimates of raters’ were recorded and analysis-of-variance tests on the mean cover estimates were investigated to see whether if there were disagreements between the raters’. In tra-correl ation coefficient was used to study the differences between the estimates. Results achieved in this work revealed that seven out of the nine analysis-of-variance tests conducted in this study have demonstrated significant difference in the mean estimates of cover (p < 0.05).
The presence of vegetation on railway tracks (amongst other issues) threatens track safety and longevity. However, vegetation inspections in Sweden (and elsewhere in the world) are currently being carried out manually. Manually inspecting vegetation is very slow and time consuming. Maintaining an even quality standard is also very difficult. A machine vision-based approach is therefore proposed to emulate the visual abilities of the human inspector. Work aimed at detecting vegetation on railway tracks has been split into two main phases. The first phase is aimed at detecting vegetation on the tracks using appropriate image analysis techniques. The second phase is aimed at detecting the rails in the image to determine the cover of vegetation that is present between the rails as opposed to vegetation present outside the rails. Results achieved in the current work indicate that the machine vision approach has performed reasonably well in detecting the presence/absence of vegetation on railway tracks when compared with a human operator.
This paper summarises the results of using a pattern recognition approach for classifying the condition of wooden railway sleepers. Railway sleeper inspections are currently done manually; visual inspection being the most common approach, with some deeper examination using an axe to judge the condition. Digital images of the sleepers were acquired to compensate for the human visual capabilities. Appropriate image analysis techniques were applied to further process the images and necessary features such as number of cracks, crack length etc have been extracted. Finally a pattern recognition and classification approach has been adopted to further classify the condition of the sleeper into classes (good or bad). A Support Vector Machine (SVM) using a Gaussian kernel has achieved good classification rate (86%) in the current case.