Online Sensor Data Research Articles

3D FEATURE POINT EXTRACTION FROM LIDAR DATA USING A NEURAL NETWORK

Accurate positioning of vehicles plays an important role in autonomous driving. In our previous research on landmark-based positioning, poles were extracted both from reference data and online sensor data, which were then matched to improve the positioning accuracy of the vehicles. However, there are environments which contain only a limited number of poles. 3D feature points are one of the proper alternatives to be used as landmarks. They can be assumed to be present in the environment, independent of certain object classes. To match the LiDAR data online to another LiDAR derived reference dataset, the extraction of 3D feature points is an essential step. In this paper, we address the problem of 3D feature point extraction from LiDAR datasets. Instead of hand-crafting a 3D feature point extractor, we propose to train it using a neural network. In this approach, a set of candidates for the 3D feature points is firstly detected by the Shi-Tomasi corner detector on the range images of the LiDAR point cloud. Using a back propagation algorithm for the training, the artificial neural network is capable of predicting feature points from these corner candidates. The training considers not only the shape of each corner candidate on 2D range images, but also their 3D features such as the curvature value and surface normal value in z axis, which are calculated directly based on the LiDAR point cloud. Subsequently the extracted feature points on the 2D range images are retrieved in the 3D scene. The 3D feature points extracted by this approach are generally distinctive in the 3D space. Our test shows that the proposed method is capable of providing a sufficient number of repeatable 3D feature points for the matching task. The feature points extracted by this approach have great potential to be used as landmarks for a better localization of vehicles.

Strategic mine planning and design: some challenges and strategies for addressing them

The authors provide their assessment of current challenges in strategic mine planning and some approaches for addressing them. Challenges covered are quantification of downstream processes and their integration into orebody models; spatial characterisation of geometallurgical variables and their integration into block models, scheduling and mine optimisation; spatial characterisation of new variables for new types of mining; flexibility in planning and design to manage risk and minimise its impact; IT infrastructure for rapid on-line data collection and processing. Most of these challenges require new types of data, variables, modelling and estimation methods. Foremost among new variable types are geometallurgical and dynamic rock mass characterisation variables. New types of data and data collection include rapid generation of very large amounts of on-line sensor data and the consequent need for rapid processing and modelling of these data. The paper includes examples of models and outputs to illustrate approaches and potential solutions.

A Self-Alignment Algorithm for SINS Based on Gravitational Apparent Motion and Sensor Data Denoising.

Initial alignment is always a key topic and difficult to achieve in an inertial navigation system (INS). In this paper a novel self-initial alignment algorithm is proposed using gravitational apparent motion vectors at three different moments and vector-operation. Simulation and analysis showed that this method easily suffers from the random noise contained in accelerometer measurements which are used to construct apparent motion directly. Aiming to resolve this problem, an online sensor data denoising method based on a Kalman filter is proposed and a novel reconstruction method for apparent motion is designed to avoid the collinearity among vectors participating in the alignment solution. Simulation, turntable tests and vehicle tests indicate that the proposed alignment algorithm can fulfill initial alignment of strapdown INS (SINS) under both static and swinging conditions. The accuracy can either reach or approach the theoretical values determined by sensor precision under static or swinging conditions.

Making Use of Online Production Data: Sequential Updating of Mineral Resource Models

The information flow along the mining value chain from exploration through resource/reserve estimation, mine planning, operations management and processing generally occurs discontinuously over long time spans. To react to deviations between produced ore and model based expectations, reconciliation exercises are performed adjusting resource models and mine planning assumptions. However, there is often a lag of weeks, months or even years. Developments over the last decade have created a flood of online data about different aspects during the production process. For example, sensor technology enables online characterization of geochemical, mineralogical and physical material characteristics. The ability to fully exploit this additional information and feed it back into resource models will open up new opportunities for improved decision making in short-term planning and operational control. This contribution introduces a new approach for sequential resource model updating utilizing online sensor data. The updating approach is based on differences between model-based predictions and sensor measurements of raw material properties. In this context, one major challenge has to be solved. Raw material streams at sensor stations often occur as a blend of material originating from multiple different extraction locations. A direct allocation of the source of differences between the model-based prediction and sensor measurements is difficult. To overcome this issue, this contribution proposes an adaption of a data assimilation-based approach for sequential model updating. The theoretical description of the method is provided followed by a demonstrative case study that investigates the performance with respect to different mine configurations.

Modeling and Prediction of Surface Water Contamination using On-line Sensor Data

Abstract Water contamination is a great disadvantage to humans and aquatic life. Maintaining the aesthetics and quality of water bodies is a priority for environmental stake holders. The water quality sensor data can be analyzed over a period of time to give an indication of pollution incidents and could be a useful forecasting tool. Here we show our initial finding from statistical analysis on such sensor data from one of the lakes of the river Lea, south of Luton. Our initial work shows patterns which will form the basis for our forecasting model.

A middleware framework for scalable management of linked streams

The Web has long exceeded its original purpose of a distributed hypertext system and has become a global, data sharing and processing platform. This development is confirmed by remarkable milestones such as the Semantic Web, Web services, social networks and mashups. In parallel with these developments on the Web, the Internet of Things (IoT), i.e., sensors and actuators, has matured and has become a major scientific and economic driver. Its potential impact cannot be overestimated–for example, in logistics, cities, electricity grids and in our daily life, in the form of sensor-laden mobile phones–and rivals that of the Web itself. While the Web provides ease of use of distributed resources and a sophisticated development and deployment infrastructure, the IoT excels in bringing real-time information from the physical world into the picture. Thus a combination of these players seems to be the natural next step in the development of even more sophisticated systems of systems. While only starting, there is already a significant amount of sensor-generated, or more generally dynamic information, available on the Web. However, this information is not easy to access and process, depends on specialised gateways and requires significant knowledge on the concrete deployments, for example, resource constraints and access protocols. To remedy these problems and draw on the advantages of both sides, we try to make dynamic, online sensor data of any form as easily accessible as resources and data on the Web, by applying well-established Web principles, access and processing methods, thus shielding users and developers from the underlying complexities. In this paper we describe our Linked Stream Middleware (LSM, http://lsm.deri.ie/), which makes it easy to integrate time-dependent data with other Linked Data sources, by enriching both sensor sources and sensor data streams with semantic descriptions, and enabling complex SPARQL-like queries across both dataset types through a novel query processing engine, along with means to mashup the data and process results. Most prominently, LSM provides (1) extensible means for real-time data collection and publishing using a cloud-based infrastructure, (2) a Web interface for data annotation and visualisation, and (3) a SPARQL endpoint for querying unified Linked Stream Data and Linked Data. We describe the system architecture behind LSM, provide details of how Linked Stream Data is generated, and demonstrate the benefits and efficiency of the platform by showcasing some experimental evaluations and the system’s interface.

Probabilistic qualitative analysis for fault detection and identification of an on-line phosphate analyzer

On-line, real-time collection of measurements remains a key challenge in water quality monitoring and control due to unknown and varying quality of on-line sensor data. Today’s data quality assessment is typically based on a comparison of sensor-based measurements and grab samples of the sampled solution taken next to the on-line analyzer and analyzed in a laboratory. In this work, internal data is used for fault detection and identification of a phosphate analyzer to inspect the measuring process itself. These internal data is shown to be information-rich with respect to the analyzer’s status. Furthermore, this information is captured well by means of a newly developed method for qualitative analysis of time series. This method was developed with global optimality in mind and therefore lends itself to a probabilistic assessment of the qualitative representation of time series.

A Middleware Framework for Scalable Management of Linked Streams

The Web has long exceeded its original purpose of a distributed hypertext system and has become a global, data sharing and processing platform. This development is confirmed by remarkable milestones such as the Semantic Web, Web services, social networks and mashups. In parallel with these developments on the Web, the Internet of Things (IoT), i.e., sensors and actuators, has matured and has become a major scientific and economic driver. Its potential impact cannot be overestimated – for example, in logistics, cities, electricity grids and in our daily life, in the form of sensor-laden mobile phones – and rival that of the Web itself. While the Web provides ease of use of distributed resources and a sophisticated development and deployment infrastructure, the IoT excels in bringing real-time information from the physical world into the picture. Thus a combination of these players seems to be the natural next step in the development of even more sophisticated systems of systems. While only starting, there is already a significant amount of sensor-generated, or more generally dynamic information, available on the Web. However, this information is not easy to access and process, depends on specialised gateways and requires significant knowledge on the concrete deployments, for example, resource constraints and access protocols. To remedy these problems and draw on the advantages of both sides, we try to make dynamic, online sensor data of any form as easily accessible as resources and data on the Web, by applying well-established Web principles, access and processing methods, thus shielding users and developers from the underlying complexities. In this paper we describe our Linked StreamMiddleware, which makes it easy to integrate time-dependent data with other Linked Data sources, by enriching both sensor sources and sensor data streams with semantic descriptions, and enabling complex SPARQL-like queries across both dataset types through a novel query processing engine, along with means to mashup the data and process results. Most prominently, LSM provides (1) extensible means for real-time data collection and publishing using a cloud-based infrastructure, (2) a Web interface for data annotation and visualisation, and (3) a SPARQL endpoint for querying unified Linked Stream Data and Linked Data. We describe the system architecture behind LSM, provide details how Linked Stream Data is generated, and demonstrate the benefits and efficiency of the platform by showcasing some experimental evaluations and the system’s interface.

Data fusion to improve trajectory tracking in a Cooperative Surveillance Multi-Agent Architecture

In this paper we present a Cooperative Surveillance Multi-Agent System (CS-MAS) architecture extended to incorporate dynamic coalition formation. We illustrate specific coalition formation using fusion skills. In this case, the fusion process is divided into two layers: (i) a global layer in the fusion center, which initializes the coalitions and (ii) a local layer within coalitions, where a local fusion agent is dynamically instantiated. There are several types of autonomous agent: surveillance–sensor agents, a fusion center agent, a local fusion agent, interface agents, record agents, planning agents, etc. Autonomous agents differ in their ability to carry out a specific surveillance task. A surveillance–sensor agent controls and manages individual sensors (usually video cameras). It has different capabilities depending on its functional complexity and limitations related to sensor-specific aspects. In the work presented here we add a new autonomous agent, called the local fusion agent, to the CS-MAS architecture, addressing specific problems of on-line sensor alignment, registration, bias removal and data fusion. The local fusion agent is dynamically created by the fusion center agent and involves several surveillance–sensor agents working in a coalition. We show how the inclusion of this new dynamic local fusion agent guarantees that, in a video-surveillance system, objects of interest are successfully tracked across the whole area, assuring continuity and seamless transitions.

Pollution load modelling in sewer systems: an approach of combining long term online sensor data with multi-objective auto-calibration schemes

Pollutant load modelling for sewer systems is state-of-the-art, especially for the estimation of discharged pollutant loads and development of sewer management strategies. However, conventionally obtained calibration data sets are often not exhaustive and have significant drawbacks. In the Graz West catchment area (Graz, Austria), continuous high-resolution long-term online measurements for discharge and pollutant concentration have been carried out since 2002.In this paper, the application of single- and multi-objective auto-calibration schemes based on evolution strategies for a deterministic hydrological pollutant load model will be discussed. Three approaches for pollutant load modelling are examined and compared: using a constant storm weather concentration and two surface accumulation-wash-off approaches with basic respectively extended wash-off equations. It is shown that the applied auto-calibration method leads to very satisfying results for both the calibration and the validation data set, and also for the dry and the storm weather runoff. Results from multi-objective calibration show better robustness in validation events than single-objective calibration. The build-up wash-off approach using the basic wash-off equation gives the best correlations between measured data and simulation results.

Implementation of an integrated on-line process surveillance and diagnostic system at the Halden reactor project: MOAS

MOAS is an integrated on-line process surveillance and diagnostic system that uses several different models for knowledge acquisition and diagnostic reasoning, such as goal-tree success-tree model, process monitor trees, and sensor failure diagnosis trees. Within these models, the knowledge of the process and its operation, including deep knowledge, like mass balance or controller algorithm, is incorporated. During an extensive review, made as part of the integrated diagnosis system project of the Halden reactor project, MOAS (Maryland Operator Advisory System) was identified as one of the most thorough systems developed thus far. MOAS encompasses diverse functional aspects that are required for an effective process disturbance management: (1) intelligent process monitoring and alarming, (2) on-line sensor data validation and sensor failure diagnosis, (3) on-line hardware (besides sensors) failure diagnosis, and (4) real-time corrective measure synthesis. The MOAS methodology was used for the NORS (Nokia Research Simulator) process at the Halden man-machine laboratory HAMMLAB of the OECD Halden reactor project. The performance tests of MOAS, implemented in G2 real-time expert system shell, show that MOAS successfully carries out its intended functions, i.e. quickly recognizing an occurring disturbance, correctly diagnosing its cause, and presenting advice on its control to the operator. The lessons learned and insights gained during the implementation and performance tests also are discussed.

Real-time diagnosis of semiconductor manufacturing equipment using a hybrid neural network expert system

This paper presents a tool for the real-time diagnosis of integrated circuit fabrication equipment. The approach focuses on integrating neural networks into an expert system. The system employs evidential reasoning to identify malfunctions by combining evidence originating from equipment maintenance history, on-line sensor data, and in-line post-process measurements. Neural networks are used in the maintenance phase of diagnosis to approximate the functional form of the failure history distribution of each component. Predicted failure rates are then converted to belief levels. For on-line diagnosis in the ease of previously unencountered faults, a CUSUM control chart is implemented on real sensor data to detect very small process shifts and their trends. For the known fault case, continuous hypothesis testing on the statistical mean and variance of the sensor data is performed to search for similar data patterns and assign belief levels. Finally, neural process models of process figures of merit (such as etch uniformity) derived from prior experimentation are used to analyze the in-line measurements, and identify the most suitable candidate among faulty input parameters (such as gas flow) to explain process shifts. A working prototype for this hybrid diagnostic system has been implemented on the Plasma Therm 700 series reactive ion etcher located in the Georgia Tech Microelectronics Research Center.

FORMENTOR: Real-time operator advisory system for loss control. Application to a petro-chemical plant

This paper describes FORMENTOR, a real-time expert system that offers on-line decision support to operators of a complex plant who are faced with unusual and potentially hazardous situations. After the overall project background has been described, the essential requirements of such a system are outlined. The paper will then focus on one of the developed pilot applications, a supervisory and decision support system for operators of an extractive distillation unit of a petro-chemical plant. The most important techniques and models used will be discussed. A number of generic and well defined tasks have been identified for a plant supervisory system like FORMENTOR to allow it to monitor on-line sensor data and detect precursor signs of abnormal plant behaviour, to diagnose the causes of the malfunction and finally to yield operational advice for managing the abnormal situation. This set of tasks include data retrieval, monitoring, diagnosing, assessing the current situation, assessing the consequences of the current situation, countermeasure planning and finally generation of advice. These generic tasks are instantiated in this particular application. In addition, main technical features of this pilot application are a functional model used to provide a global overview of the plant state and to guide the action planning process, a multi-layered structural model used to represent the functional components of the plant in a hierarchical way, and the use of two distinct but complementary reasoning modules for diagnosis, both based on this multi-level model. Of the two diagnostic reasoning modules one, the Heuristic Causal Reasoning module, identifies known problems, looks up the known solutions, and then adapts the implementation of these high-level solutions to the plant state. The other, the Model-Based Reasoning module, uses a more sophisticated model-based reasoning to find the root causes of a disturbance. The developed FORMENTOR system has been installed on an extractive distillation plant operated by BP-Chemicals Ltd. Its performance in responding to real plant upsets has been evaluated. The main results of this testing campaign are also reported.

A model-based approach to on-line process disturbance management: The models

Abstract A methodology is proposed which can be used to design real-time expert systems for on-line process disturbance management. This methodology encompasses diverse functional aspects that are required for an effective process disturbance management: (1) intelligent process monitoring and alarming, (2) on-line sensor data validation and sensor failure diagnosis, (3) on-line hardware (besides sensors) failure diagnosis, and (4) real-time corrective measure synthesis. Accomplishment of these functions is made possible through the integrated application of the various models: goal-tree success-tree, process monitor tree, sensor failure diagnosis, and hardware failure diagnosis models. This paper presents and discusses the various models along with the overall algorithm of the methodology. The application of the methodology to a target process, a typical main feedwater system of a nuclear power plant which employs a complex control mechanism, will be presented in a companion paper.

Man-Machine Interface Issues in the Design of An Automated Decision System

This paper describes issues in building an effective interface between an automated decision system and the human user of that system's output. Designing an interface so that the user can understand and evaluate the output is an important issue, especially with advances in technology that increase the potential for decision automation. The specific issues addressed during the design application discussed here are: 1) the development of an object display to present the output of an automated decision system when the decision can be only partially automated and 2) presenting the logic used in automated decisions to a human user. The application is a decision system developed to help nuclear power plant operators determine what response should be made during severe emergencies. The logic is executed by a computer to produce real time output using on-line sensor data.