The paper discusses the sensor selection problem in estimating spatial fields. It is demonstrated that selecting a subset of sensors depends on modelling spatial processes. It is first proposed to exploit Gaussian process (GP) to model a univariate spatial field and multivariate GP (MGP) to jointly represent multivariate spatial phenomena. A Mat´ern cross covariance function is employed in the MGP model to guarantee its cross-covariance matrices to be positive semi-definite. We then consider two corresponding univariate and multivariate sensor selection problems in effectively monitoring multiple spatial random fields. The sensor selection approaches were implemented in the real-world experiments and their performances were compared. Difference of results obtained by the univariate and multivariate sensor selection techniques is insignificant; that is, either of the methods can be efficiently used in practice.

Sensors help robots perceive their environment and localize themselves. Determining a robot’s location requires a range of sensing systems. Depending on accuracy criteria and navigation conditions, robot localization sensors can differ. Common sensors for robot localization include encoders, GPS, cameras, LIDARs, and IMUs. Traditional sensors are not capable enough in changing environments and uneven terrain. In this paper, we propose a method based on deep learning to use the subsurface features obtained through a Ground Penetrating Radar (GPR) to estimate the odometry of a robot. This proposed method does not rely on visual features or the distance gathered from wheel encoders. The proposed approach was evaluated on a publicly available dataset, and the evaluation results show that the proposed method can be used for robot localization without the need for odometry from wheel encoders.

Water utilities across the globe are concerned with the inspection and replacement of buried metallic water pipes due to corrosion-related structural damages. Internal pipe linings are commonly used as a renewal method to improve structural strength as they are regarded to be a less expensive alternative to costly and time-consuming pipe replacements. However, linings are also prone to failure as well. Therefore, water authorities regularly monitor lining performance, where defect evolution over a long period of time is an important parameter to note. It requires an accurate in-pipe robot localization technology. In this article, we propose a novel method for in-pipe robot localization and tag mapping that uses battery-free UHF-RFID sensor wireless signals. It utilizes a signal mapping approach in combination with a tailored pose-graph simultaneous localization and mapping algorithm. Evaluation results of a field-extracted pipe sample from Sydney Water’s distribution network show the proposed approach is capable of localizing the robot within 2.5cm accuracy in a 50m equivalent pipe with an unknown UHF-RFID distribution. The proposed approach outperformed other reported similar work in the literature.

Optimal sensor placement is an important problem to look at. This problem becomes all the more relevant nowadays due to advancements in infrastructure monitoring robotic technologies including underground sensing. While there are multiple ways to solve optimal sensor placement problems, one of the most generic methods available is Bayesian Optimization and its variants. In this paper, we present a simple benchmark- like formulation for exploiting Gaussian Process uncertainty for sensor placement to measure a scalar field.

A reliable robotic localization method is required for comparing three-dimensional pipe maps obtained via laser scans at various times for accurately monitoring the evolution of internal pipe surface defects. Existing robotic localization methods have limitations when visual features vanish due to changes in the pipe environment or when encoder data becomes highly uncertain due to long-distance robotic traverses. To address this issue, we leverage battery-free ultra-high frequency radio frequency identification (UHF-RFID) sensors for transmitting wireless signals to a two-antenna reader integrated mobile robotic system. Although there are literature on the investigation of UHFRFID behaviors and their applications in indoor environments, analysis of the same for in-pipe scenarios was not well studied. In this paper, we evaluate the UHF-RFID sensor signals inside a field extracted pipeline. Firstly, we examine the UHF-RFID sensor signal patterns through repeated robotic scans. Secondly, we examine how the placement of UHF-RFID reader antennas affects the transmission of UHF-RFID sensor signals, as well as we study the effects of robotic traverse direction and speed on the UHF-RFID wireless signals. Finally, we examine whether identical UHF-RFID sensors generate the same pattern when placed in a pipeline. Overall, the experimental evaluation demonstrates that the use of two-antenna UHF-RFID readers can ameliorate the capabilities of robotic localization in the pipeline.

Underground water pipes are important to any country's infrastructure. Overtime, the metallic pipes are prone to corrosion, which can lead to water leakage and pipe bursts. In order to prolong the service life of those assets, water utilities in Australia apply protective pipe linings. Long-term monitoring and timely intervention are crucial for maintaining those lining assets. However, the water utilities do not possess the comprehensive technology to achieve it. The main reasons for lacking such technology are the unavailability of sensors and accurate robot localization technologies. Feature based localization methods such as SLAM has limited use as the application of liners alters the features and the environment. Encoder based localization is not accurate enough to observe the evolution of defects over a long period of time requiring unique defect correspondence. This motivates us to explore accurate contact-less and wireless based localization methods. We propose a cost-effective localization method using UHFRFID signals for robot localization inside pipelines based on Gaussian process combined particle filter. Experiments carried out in field extracted pipe samples from the Sydney water pipe network show that using the RSSI and Phase data together in the measurement model with particle filter algorithm improves the localization accuracy up to 15 centimeters precision.

Identification of static nonlinear elements (i.e., nonlinear elements whose outputs depend only on the present value of inputs) is crucial for the success of system identification tasks. Identification of static nonlinear elements though can pose several challenges. Two of the main challenges are: (1) mathematical models describing the elements being unknown and thus requiring black-box identification; and (2) collection of sufficiently informative measurements. With the aim of addressing the two challenges, we propose in this paper a method of predetermining informative measurement points offline (i.e., prior to conducting experiments or seeing any measured data), and using those measurements for online model calibration. Since we deal with an unknown model structure scenario, a high order polynomial model is assumed. Over fit and under fit avoidance are achieved via checking model convergence via an iterative means. Model dependent information maximization is done via a D-optimal design of experiments strategy. Due to experiments being designed offline and being designed prior to conducting measurements, this method eases off the computation burden at the point of conducting measurements. The need for in-the-loop information maximization while conducting measurements is avoided. We conclude by comparing the proposed D-optimal design method with a method of in-the-loop information maximization and point out the pros and cons. The method is demonstrated for the single-input-single-output (SISO) static nonlinear element case. The method can be extended to MISO systems as well.

The paper addresses the multimodal sensor selection problem where selected collocated sensor nodes are employed to effectively monitor and efficiently predict multiple spatial random fields. It is first proposed to exploit multivariate Gaussian processes (MGP) to model multiple spatial phenomena jointly. By the use of the Matern cross-covariance function, cross covariance matrices in the MGP model are sufficiently positive semi-definite, concomitantly providing efficient prediction of all multivariate processes at unmeasured locations. The multimodal sensor selection problem is then formulated and solved by an approximate algorithm with an aim to select the most informative sensor nodes so that prediction uncertainties at all the fields are minimized. The proposed approach was validated in the real-life experiments with promising results.

Microbial corrosion is considered the main reason for multi-billion dollar sewer asset degradation. Sewer pipe surface temperature is a vital parameter for predicting the micro-biologically induced concrete corrosion. Due to this important measure, a surface temperature sensor suite was recently developed and tested in an aggressive sewer environment. The sensors can fail and they may also put offline during the period of scheduled maintenance. In such situations, time series forecasting of sensor data can be an alternative measure for the operators managing the sewer network. In this regard, this paper focuses on the short-term forecasting of sensor measurements. The evaluation was carried out by forecasting the sensor measurements for different time periods and evaluated with different forecasting models. The ETS model leads to high short-term forecasting accuracy and the ARIMA model leads to high long-term forecasting accuracy. The models were evaluated on real data captured in a Sydney sewer

Classification has become a vital task in modern machine learning and Artificial Intelligence applications, including smart sensing. Numerous machine learning techniques are available to perform classification. Similarly, numerous practices, such as feature selection (i.e., selection of a subset of descriptor variables that optimally describe the output), are available to improve classifier performance. In this paper, we consider the case of a given supervised learning classification task that has to be performed making use of continuous-valued features. It is assumed that an optimal subset of features has already been selected. Therefore, no further feature reduction, or feature addition, is to be carried out. Then, we attempt to improve the classification performance by passing the given feature set through a transformation that produces a new feature set which we have named the “Binary Spectrum”. Via a case study example done on some Pulsed Eddy Current sensor data captured from an infrastructure monitoring task, we demonstrate how the classification accuracy of a Support Vector Machine (SVM) classifier increases through the use of this Binary Spectrum feature, indicating the feature transformation’s potential for broader usage.

Hyper-parameter optimization is an essential task in the use of machine learning techniques. Such optimizations are typically done starting with an initial guess provided to hyperparameter values followed by optimization (or minimization) of some cost function via gradient-based methods. The initial values become crucial since there is every chance for reaching local minimums in the cost functions being minimized, especially since gradient-based optimizing is done. Therefore, initializing hyper-parameters several times and repeating optimization to achieve the best solutions is usually attempted. Repetition of optimization can be computationally expensive when using techniques like Gaussian Process (GP) which has an O(n3) complexity, and not having a formal strategy to initialize hyperparameter values is an additional challenge. In general, reinitialization of hyper-parameter values in the contexts of many machine learning techniques including GP has been done at random over the years; some recent developments have proposed some initialization strategies based on the optimization of some meta loss cost functions. To simplify this challenge of hyperparameter initialization, this paper introduces a data-dependent deterministic initialization technique. The specific case of the squared exponential kernel-based GP regression problem is focused on, and the proposed technique brings novelty by being deterministic as opposed to random initialization, and fast (due to the deterministic nature) as opposed to optimizing some form of meta cost function as done in some previous works. Although global suitability of this initialization technique is not proven in this paper, as a preliminary study the technique’s effectiveness is demonstrated via several synthetic as well as real data-based nonlinear regression examples, hinting that the technique may have the effectiveness for broader usage.

Systems containing linear first-order dynamics and static nonlinear elements (i.e., nonlinear elements whose outputs depend only on the present value of inputs) are often encountered; for example, certain automobile engine subsystems. Therefore, system identification of static nonlinear elements becomes a crucial component that underpins the success of the overall identification of such dynamical systems. In relation to identifying such systems, we are often required to identify models in differential equation form, and consequently, we are required to describe static nonlinear elements in the form of functions in time domain. Identification of such functions describing static elements is often a black-box identification exercise; although the inputs and outputs are known, correct mathematical models describing the static nonlinear elements may be unknown. Therefore, with the aim of obtaining computationally efficient models, calibrating polynomial models for such static elements is often attempted. With that approach comes several issues, such as long time requirements to collect adequate amounts of measurements to calibrate models, having to test different models to pick the best one, and having to avoid models over-fitting to noisy measurements. Given that premise, this paper proposes an approach to tackle some of those issues. The approach involves collecting measurements based on an uncertainty-driven Active Learning scheme to reduce time spent on measurements, and simultaneously fitting smooth models using Gaussian Process (GP) regression to avoid over-fitting, and subsequently picking best fitting polynomial models using GP-regressed smooth models. The principles for the single-input-single-output (SISO) static nonlinear element case are demonstrated in this paper through simulation. These principles can easily be extended to MISO systems.