Expectation Maximization Technique Research Articles

Joint PAPR reduction with channel estimation for OFDM-based VLC systems

Abstract This paper presents a new combination of channel estimation (CE) methods and peak-to-average power ratio (PAPR) reduction techniques for orthogonal frequency division multiplexing (OFDM)-based visible light communication (VLC) systems. We first propose iterative clipping and filtering to minimize the PAPR of optical OFDM signals. Then, we introduce a computationally efficient, high-performance channel estimation approach using block-type pilots with modified clipped signals. The proposed method integrates clipping with three distinct CE techniques, resulting in both significant PAPR reduction and enhanced bit error rate (BER) performance. Among the CE methods evaluated, the expectation-maximization (EM) technique provides better BER results compared to other approaches. Simulation results confirm that the combined use of clipping and advanced CE methods leads to notable improvements in PAPR reduction and BER performance for OFDM-based VLC systems.

Bayesian expectation maximization-maximization for robust estimation in proton exchange membrane fuel cells: A comparative study

In this paper, we present a novel approach based on Bayesian Expectation Maximization-Maximization (BEMM) to address this challenge. Unlike traditional optimization methods, which may struggle with high-dimensional and nonlinear optimization problems, BEMM offers a robust framework that combines the benefits of Bayesian inference with the flexibility of expectation maximization and maximization techniques. By iteratively updating parameter estimates based on observed data and maximizing the likelihood of the model, BEMM effectively navigates the solution space to converge on accurate estimates of the unspecified variables in Proton Exchange Membrane Fuel Cells (PEMFCs)models. Through extensive experimentation and comparison with other metaheuristic techniques, including Arithmetic Optimization Algorithm (AOA), Gravitational Search Algorithm (GSA), Flower Pollination Algorithm (FPA), and Biogeography-Based Optimization (BBO), we demonstrate the superior performance of our BEMM approach. Our results show that BEMM outperforms these alternative methods in terms of precision, convergence speed, and stability across various scenarios involving different numbers of unspecified variables. The implications of our findings are significant for both researchers and practitioners in the field of PEMFC modeling and optimization. By providing a more efficient and reliable method for estimating model parameters, our approach can facilitate more accurate predictions of PEMFC performance, leading to better-informed decision-making in the design, operation, and optimization of PEMFC systems. Furthermore, the robustness and versatility of BEMM make it well-suited for a wide range of optimization problems beyond PEMFC modeling, highlighting its potential impact across various domains of engineering and scientific research. In the sensitivity analysis, as the population size increases from 10 to 40, there is a significant improvement in solution quality by approximately 100 %. However, beyond a population size of 40, the marginal gains diminish, with only marginal improvements of less than 1 % observed despite further increases in population size up to 200.

Crop recommendation and forecasting system for Maharashtra using machine learning with LSTM: a novel expectation-maximization technique

Agriculture in Maharashtra has immense importance in India, acting as the back-bone of the economy and a primary livelihood source for a significant population. Being the third largest state in India, Maharashtra has a high scale crop production in the country which also has an important impact on the economy. Initially the study focus on developing predictive models that guide farmers in selecting suitable crops for the divisions in the state of Maharashtra. This study presents a Crop Recommendation System (CRS) designed to support Maharashtra’s agricultural sector by utilizing a comprehensive dataset from 2001 to 2022 provided by the India Meteorological Department. This study helps in improvising technical efficiency and productivity of the farmers. Harvesting crops in optimal condition can help to produce efficient harvest hence the research concentrates on providing best crop recommendation system (CRS) with the help of Machine Learning and Deep Learning techniques. The data, enhanced for accuracy using expectation-maximization optimization, underpins predictive models that guide crop selection. EM contributes to a more robust and reliable dataset for subsequent analyses and modeling by iterative estimating and updating missing values based on probabilistic expectations. Key findings show that the Random Forest algorithm excels in predicting suitable crops with 92% accuracy. Further precision is achieved through a Long Short-Term Memory network forecasting weather patterns three months ahead, accommodating temporal data variations. Subsequently, the proposed system leverages these forecasts to recommend five ideal crops per division within Maharashtra, aiding farmers’ decision-making and adapting to regional climatic conditions. A supplementary crop calendar offers monthly district-specific planting guidance. An intuitive Graphical User Interface delivers this information effectively, ensuring practical and informed agricultural choices across the state. In essence, the study provides an innovative tool for enhancing economic stability and sustenance in Maharashtra through technology-driven agriculture recommendations aligned with future weather expectations.

A comprehensive evaluation of OPTICS, GMM and K-means clustering methodologies for geochemical anomaly detection connected with sample catchment basins

The process of data-driven clustering to uncover geochemical anomalies linked to sample catchment basins (SCBs) includes a comprehensive framework to discern areas exhibiting unique geochemical attributes within a specified study area. The Ordering Points to Identify the Clustering Structure (OPTICS) method can serve as a robust methodology for detecting geochemical anomalies in SCBs. This is attributed to its capacity to effectively manage varying cluster densities, adaptively identify cluster numbers, exhibit resilience to noise, and display minimum sensitivity to parameters. A comparison was conducted in this research between the outcomes of the OPTICS clustering algorithm and two traditional clustering techniques, namely the Gaussian Mixture Model (GMM) and K-means clustering. In the following, the Expectation-Maximization (EM) technique is employed to train the GMM for clustering. Moreover, the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) as two validate statistical metrics implemented to ascertain the optimal number of components (clusters) belong to the GMM. It should be noted that the effectiveness of the clustering algorithms was further assessed using the Calinski-Harabasz (CH) index and the success-rate curves. OPTICS, a density-based clustering approach, was confirmed to be more effective than K-means and GMM for identifying MVT PbZn anomalies in Varcheh district, western Iran. Furthermore, the specified anomalies show a geo-spatial correspondence with the geological facts, and it has been observed that strong anomalies are more discoverable in close proximity to MVT PbZn occurrences. This work suggests a novel anomaly detection approach based on OPTICS, which exhibits superior performance and data-modeling efficiency. The main emphasis is on effectively distinguishing geochemical anomalies from sample data originating from populations with uncertain distributions.

Bayesian two-stage structural identification with equivalent formulation and EM algorithm

For structural model identification using a Bayesian two-stage approach, modal properties (e.g., natural frequencies) are first extracted from measured data in Stage I and then used for determining structural properties (e.g., stiffness) in Stage II. With sufficient data that allow the problem to be globally identifiable, the computational problem often reduces to optimizing a ‘measure-of-fit’ function to yield the ‘most probable value’ (MPV) that informs the ‘best estimate’, and determining the Hessian of the function to yield the posterior covariance matrix that informs the remaining uncertainty. Recent developments deal with the presence of model error, and the increasing complexity calls for proper computational strategy. In this spirit, assuming Gaussian model error, this work develops a hypothetical yet mathematically equivalent formulation for the two-stage problem, which facilitates development of effective algorithms for MPV using Expectation-Maximization techniques. By treating hypothetically the MPV of modal properties in Stage I as ‘data’ and model error as latent variables, the Q-function in the M−step can be expressed as a sum of two terms that can be optimized with respect to structural parameters and model error parameters. The optimization of structural parameters reduces to that of a Stage II problem without model error, for which existing algorithms can be applied. Analytical expressions are also derived for computing the posterior covariance matrix. The proposed methodology is investigated with synthetic data for model errors associated with sensor misalignment and model simplification, lab data for the effect of number of modes and measured degrees of freedom (DoFs), and field data for reality tests.

Unsupervised Classification under Uncertainty: The Distance-Based Algorithm

This paper presents a method for unsupervised classification of entities by a group of agents with unknown domains and levels of expertise. In contrast to the existing methods based on majority voting (“wisdom of the crowd”) and their extensions by expectation-maximization procedures, the suggested method first determines the levels of the agents’ expertise and then weights their opinions by their expertise level. In particular, we assume that agents will have relatively closer classifications in their field of expertise. Therefore, the expert agents are recognized by using a weighted Hamming distance between their classifications, and then the final classification of the group is determined from the agents’ classifications by expectation-maximization techniques, with preference to the recognized experts. The algorithm was verified and tested on simulated and real-world datasets and benchmarked against known existing algorithms. We show that such a method reduces incorrect classifications and effectively solves the problem of unsupervised collaborative classification under uncertainty, while outperforming other known methods.

Speech Enhancement Based on Approximate Message Passing

Abstract: A novel speech improvement technique based on the approximate message passing (AMP) is adopted to get beyond the drawbacks of traditional speech enhancement methods, such as inaccurate voice activity detector (VAD) and noise estimation. To eliminate or muffle the noise from the distorted speech, AMP takes advantage of the difference between speech and noise sparsity. The AMP method is used to effectively rebuild clean speech for speech augmentation. More specifically, the prior probability distribution of the speech sparsity coefficient is represented by a Gaussian model. The expectation maximization (EM) technique provides excellent learning of the hyper-parameters of the prior model. We use the k-nearest neighbor (k-NN) approach to learn the sparsity while taking into account the correlation between the speech coefficients in neighboring frames. Additionally, computational simulations are used to verify the proposed algorithm, which outperforms the three methods which are Kalman filter, principal component analysis (PCA), independent component analysis (ICA) under a variety of signal to noise ratios and compression ratios (SNRs).

Counter-checking uncertainty calculations in Bayesian operational modal analysis with EM techniques

Bayesian operational modal analysis makes inference about the modal properties (e.g., natural frequency, damping ratio) of a structure using ‘output-only’ ambient vibration data. With sufficient data in applications, the posterior probability density function (PDF) of modal properties can be approximated by a Gaussian PDF, whose covariance matrix is given by the inverse of the Hessian of negative log-likelihood function (NLLF) at the most probable value. Existing methodologies for computing the Hessian are based on semi-analytical formulae that offer an efficient and reliable means for applications. Inevitably, their computer coding can be involved, e.g., a mix of variables with different sensitivities, singularity of Hessian due to constraints. In the absence of analytical or numerically ‘exact’ result for benchmarking, computer code verification during development stage is also non-trivial. Currently, finite difference method is often used as the only and last resort for verification, although there are also difficulties in, e.g., the choice of step size, and criterion for comparison/convergence. Motivated by these, this work explores an identity in the theory of Expectation-Maximisation (EM) algorithm to provide an alternative means for evaluating the Hessian of NLLF. Such identity allows one to evaluate the Hessian by means of Monte Carlo simulation, averaging over random samples of hidden variables. While the existing semi-analytical approach is still preferred for Hessian calculations in applications for its high definitive accuracy and speed, the proposed Monte Carlo solution offers a convenient means for counter-checking during code development. Theoretical implications of the identity will be discussed and numerical examples will be given to illustrate implementation aspects.

Reducing latency in optical 6G waveforms using joint detection and channel estimation techniques

Abstract The pursuit of ultra-low latency is a pivotal goal in advancing the capabilities of optical non-orthogonal multiple access sixth generation (O-NOMA-6G) waveforms. This study explores the application of maximum likelihood (ML) and expectation-maximization (EM) techniques to mitigate latency in optical communication systems. ML offers direct parameter estimation for rapid symbol detection, while EM addresses latency through iterative estimation of hidden variables and parameters. By leveraging the benefits of both techniques, this research proposes novel latency reduction approaches in optical 6G. The investigation encompasses theoretical analysis, simulation, and performance evaluation under Rician and Rayleigh channel conditions for different parameters such as bit error rate (BER), power spectral density (PSD) and peak to average power density (PSD). Simulation results demonstrate that ML and EM effectively reduce latency, and enable seamless integration of time-sensitive applications in optical 6G networks as compared with the conventional ML and ML methods. The outcomes of this study provide valuable insights into throughput and PSD enhancement contributing to the realization of ultra-responsive and O-NOMA 6G waveform.

Data-Aided CSI Estimation Using Affine-Precoded Superimposed Pilots in Orthogonal Time Frequency Space Modulated MIMO Systems

An orthogonal affine-precoded superimposed pilot (AP-SIP)-based architecture is developed for the cyclic prefix (CP)-aided single input single output (SISO) and multiple input multiple output (MIMO) orthogonal time frequency space (OTFS) systems relying on arbitrary transmitter-receiver (Tx-Rx) pulse shaping. The data and pilot symbol matrices are affine-precoded and superimposed in the delay Doppler (DD)-domain followed by the development of an end-to-end DD-domain relationship for the input-output symbols. At the receiver, the decoupled pilot and data symbol are extracted by employing orthogonal precoder matrices, which eliminates the mutual interference. Furthermore, a novel pilot-aided Bayesian learning (PA-BL) technique is conceived for the channel state information (CSI) estimation of SISO OTFS systems based on the expectation-maximization (EM) technique. Subsequently, a data-aided Bayesian learning (DA-BL)-based joint CSI estimation and data detection technique is proposed, which beneficially harnesses the estimated data symbols for improved CSI estimation. In this scenario our sophisticated data detection rule also integrates the CSI uncertainty of channel estimation into our the linear minimum mean square error (LMMSE) detectors. The AP-SIP framework is also extended to MIMO OTFS systems, wherein the DD-domain input matrix is affine-precoded for each transmit antenna (TA). Then an EM algorithm-based PA-BL scheme is derived for simultaneous row-group sparse CSI estimation for this system, followed also by our data-aided DA-BL scheme that performs joint CSI estimation and data detection. Moreover, the Bayesian Cramer-Rao bounds (BCRBs) are also derived for both SISO as well as MIMO OTFS systems. Finally, simulation results are presented for characterizing the performance of the proposed CSI estimation techniques in a range of typical settings along with their bit error rate (BER) performance in comparison to an ideal system having perfect CSI.

Intelligent Modeling for Batch Polymerization Reactors with Unknown Inputs

While system identification methods have developed rapidly, modeling the process of batch polymerization reactors still poses challenges. Therefore, designing an intelligent modeling approach for these reactors is important. This paper focuses on identifying actual models for batch polymerization reactors, proposing a novel recursive approach based on the expectation-maximization algorithm. The proposed method pays special attention to unknown inputs (UIs), which may represent modeling errors or process faults. To estimate the UIs of the model, the recursive expectation-maximization (EM) technique is used. The proposed algorithm consists of two steps: the E-step and the M-step. In the E-step, a Q-function is recursively computed based on the maximum likelihood framework, using the UI estimates from the previous time step. The Kalman filter is utilized to calculate the estimates of the states using the measurements from sensor data. In the M-step, analytical solutions for the UIs are found through local optimization of the recursive Q-function. To demonstrate the effectiveness of the proposed algorithm, a practical application of modeling batch polymerization reactors is presented. The performance of the proposed recursive EM algorithm is compared to that of the augmented state Kalman filter (ASKF) using root mean squared errors (RMSEs). The RMSEs obtained from the proposed method are at least 6.52% lower than those from the ASKF method, indicating superior performance.

Imaging quality improvement of ghost imaging under high correlation coefficient between speckle patterns

High correlation coefficient between speckle patterns (CCSP) results in poor imaging quality in ghost imaging (GI), especially for long-distance imaging. In this paper, we derive the analytic formula of CCSP generated by RGG and verify its effectiveness through experiments. Aiming at the degradation of imaging quality caused by high average correlation coefficient, we propose a new subset division scheme based on ordered subset expectation maximization technique to improve the quality of ghost-image. The experimental and simulation results demonstrate that our scheme can get better imaging quality under the same average correlation coefficient when compared with the commonly used GI algorithms. Even when these algorithms fail to reconstruct the target due to high average correlation coefficient, our scheme still obtains good imaging results.

Intelligent Clustering Techniques for the Reduction of Chemicals in Water Treatment Plants

Currently, the use of intelligent models for decision making in the water treatment process is very important, as many plants support their implementation with the aim of obtaining economic, social, and environmental gains. Nevertheless, for these systems to be properly modeled, the data should be carefully selected so that only those that represent good operating practices are used. Thus, this study proposes an approach for identifying water quality and operational scenarios using the expectation maximisation (EM) and self-organising maps (SOMs) techniques when using data from a water treatment plant. The results showed that both techniques were able to identify quantities of different scenarios, some similar and others different, allowing for the evaluation of differences in a robust way. The EM technique resulted in fewer scenarios when compared with the SOMs technique, including in the cluster selection process. The results also indicated that an intelligent model can be trained with data from the proposed clustering, which improves its prediction capacity under different operating conditions; this can lead to savings in chemical product usage and less waste generation throughout the water treatment process, which is in good agreement with cleaner production practices.

Unsupervised classification for uncertain varying responses: The wisdom-in-the-crowd (WICRO) algorithm

This paper addresses the problem classification of instances/questions based on the opinions (classes) provided by anonymous agents. The solution aggregates the agents’ classifications, aiming to obtain as close as possible to an unknown correct classification. However, the agents’ fields or domains of competence and their levels of expertise are unknown and can vary extensively. Many popular classification algorithms address such a problem by following a “wisdom-of-the-crowd” approach while using different voting methods and expectation–maximization techniques. These algorithms lead to correct classifications when the majority of the agents are experts, thus classifying the instances correctly. However, they often result in erroneous classification when only a small subset of the agents are indeed correct. Moreover, these algorithms often assume a fixed set of classes for all instances. This study presents a fast (one-pass) classification algorithm that can estimate the unknown agents’ expertise level and aggregates their classifications accordingly, even when these are obtained from different questionnaires; thus, when the instances are not necessarily classified to a fixed set of classes. The proposed algorithm finds the experts and the nonexpert agents for each question by analyzing the distance between them. The algorithm identifies the expert agents for each instance and then classifies them accordingly. The suggested algorithm is validated and compared against known methods by using both simulated datasets and real-world datasets collected from various sources. The obtained results clearly demonstrate the effectiveness and advantages of the proposed method.

Discovery of Intentional Self-Harm Patterns from Suicide and Self-Harm Surveillance Reports.

The purpose of this study was to identify patterns of self-harm risk factors from suicide and self-harm surveillance reports in Thailand. This study analyzed data from suicide and self-harm surveillance reports submitted to Khon Kaen Rajanagarindra Psychiatric Hospital, Thailand. The process of identifying patterns of self-harm risk factors involved: data preprocessing (namely, data preparation and cleaning, missing data management using listwise deletion and expectation-maximization techniques, subgrouping factors, determining the target factors, and data correlation for learning); classifying the risk of self-harm (severe or mild) using 10-fold cross-validation with the support vector machine, random forest, multilayer perceptron, decision tree, k-nearest neighbors, and ensemble techniques; data filtering; identifying patterns of self-harm risk factors using 10-fold cross-validation with the classification and regression trees (CART) technique; and evaluating patterns of self-harm risk factors. The random forest technique was most accurate for classifying the risk of self-harm, with specificity, sensitivity, and F-score of 92.84%, 93.12%, and 91.46%, respectively. The CART technique was able to identify 53 patterns of self-harm risk, consisting of 16 severe self-harm risk patterns and 37 mild self-harm risk patterns, with an accuracy of 92.85%. In addition, we discovered that the type of hospital was a new risk factor for severe selfharm. The procedure presented herein could identify patterns of risk factors from self-harm and assist psychiatrists in making decisions related to self-harm among patients visiting hospitals in Thailand.

Data-Driven Fault Detection in Industrial Batch Processes Based on a Stochastic Hybrid Process Model

This paper presents a novel fault detection approach for industrial batch processes. The batch processes under consideration are characterized by the interaction between discrete system modes and non-stationary continuous dynamics. Therefore, a stochastic hybrid process model (SHPM) is introduced, where process variables are modeled as time-variant Gaussian distributions, which depend on hidden system modes. Transitions between the system modes are assumed to be either autonomous or to be triggered by observable events such as on/off signals. The model parameters are determined from training data using expectation-maximization techniques. A new fault detection algorithm is proposed, which assesses the likelihoods of sensor signals on the basis of the stochastic hybrid process model. Evaluation of the proposed fault detection system has been conducted for a penicillin production process, with the results showing a significant improvement over the existing baseline methods.Note to Practitioners—Automatic fault detection makes it possible to limit the effects of faults by taking countermeasures at an early stage. In this work, a data-driven fault detection method for industrial batch processes is proposed, in which the underlying process model is learned from training data. The proposed fault detection system can be used for various industrial batch processes without the need for complex and error-prone manual configuration. In contrast to many other data-driven approaches such as neural networks, only a few process cycles are required to create a robust process model. It should be noted that in data-driven fault detection methods, the training data should cover a large part of the process states that occur during error-free process cycles. The developed method is therefore particularly suitable for cyclical processes, which, however, can have alternative process paths and variability between the process cycles.

Joint Rigid Registration of Multiple Generalized Point Sets With Anisotropic Positional Uncertainties in Image-Guided Surgery

In medical image analysis (MIA) and computer-assisted surgery (CAS), aligning two multiple point sets (PSs) together is an essential but also a challenging problem. For example, rigidly aligning multiple point sets into one common coordinate frame is a prerequisite for statistical shape modelling (SSM). Accurately aligning the pre-operative space with the intra-operative space in CAS is very crucial to successful interventions. In this article, we formally formulate the multiple generalized point set registration problem (MGPSR) in a probabilistic manner, where both the positional and the normal vectors are used. The six-dimensional vectors consisting of both positional and normal vectors are called as generalized points. In the formulated model, all the generalized PSs to be registered are considered to be the realizations of underlying unknown hybrid mixture models (HMMs). By assuming the independence of the positional and orientational vectors (i.e., the normal vectors), the probability density function (PDF) of an observed generalized point is computed as the product of Gaussian and Fisher distributions. Furthermore, to consider the anisotropic noise in surgical navigation, the positional error is assumed to obey a multi-variate Gaussian distribution. Finally, registering PSs is formulated as a maximum likelihood (ML) problem, and solved under the expectation maximization (EM) technique. By using more enriched information (i.e., the normal vectors), our algorithm is more robust to outliers. By treating all PSs equally, our algorithm does not bias towards any PS. To validate the proposed approach, extensive experiments have been conducted on surface points extracted from CT images of (i) a human femur bone model; (ii) a human pelvis bone model. Results demonstrate our algorithm’s high accuracy, robustness to noise and outliers. Note to Practitioners—This paper was motivated by solving the problem of registering two or more PSs. Most existing registration approaches use only the positional information associated with each point, and thus lacks robustness to noise and outliers. Three significant improvements are brought by our proposed approach. First, the normal vectors that can be extracted from the point sets are utilized in the registration. Second, the positional error distribution is assumed to be anisotropic and inhomogeneous. Third, all the PSs to be registered are treated equally that means no PS is considered as the model one. The registration problem is cast into a maximum likelihood (ML) problem and solved under the expectation maximization (EM) framework. We have demonstrated through extensive experiments that the proposed registration approach achieves significantly improved accuracy, robustness to noise and outliers. The algorithm is particularly suitable for biomedical applications involving the registration procedures, such as image-guided surgery.

Factors impacting customer purchase intention of smart home security systems: Social data analysis using machine learning techniques

The Internet of Things (IoTs) market for consumers includes a huge variety of smart home devices. This research is conducted to investigate what factors influence customer purchase of smart home security systems. A new model is developed and the data collection is performed from the customers' reviews on the smart home security systems on Amazon.com. Totally, 12,678 reviews were collected to investigate customers' preferences in purchasing smart home security systems. To analyze the collected data, we develop a new method using text mining, unsupervised clustering, and a neuro-fuzzy system. The factors are discovered from textual reviews using the Latent Dirichlet Allocation (LDA) technique. To segment the customers' preferences the Expectation-Maximization (EM) technique was used. ANFIS technique was then used to find the relationship between the factors and customers' purchase intention. It was demonstrated that the use of text mining can be effective in the analysis of customers’ preferences from online customer reviews. The results of the data analysis are provided and discussed.

Missing Data Imputation in the Internet of Things Sensor Networks

The Internet of Things (IoT) has had a tremendous impact on the evolution and adoption of information and communication technology. In the modern world, data are generated by individuals and collected automatically by physical objects that are fitted with electronics, sensors, and network connectivity. IoT sensor networks have become integral aspects of environmental monitoring systems. However, data collected from IoT sensor devices are usually incomplete due to various reasons such as sensor failures, drifts, network faults and various other operational issues. The presence of incomplete or missing values can substantially affect the calibration of on-field environmental sensors. The aim of this study is to identify efficient missing data imputation techniques that will ensure accurate calibration of sensors. To achieve this, we propose an efficient and robust imputation technique based on k-means clustering that is capable of selecting the best imputation technique for missing data imputation. We then evaluate the accuracy of our proposed technique against other techniques and test their effect on various calibration processes for data collected from on-field low-cost environmental sensors in urban air pollution monitoring stations. To test the efficiency of the imputation techniques, we simulated missing data rates at 10–40% and also considered missing values occurring over consecutive periods of time (1 day, 1 week and 1 month). Overall, our proposed BFMVI model recorded the best imputation accuracy (0.011758 RMSE for 10% missing data and 0.169418 RMSE at 40% missing data) compared to the other techniques (kNearest-Neighbour (kNN), Regression Imputation (RI), Expectation Maximization (EM) and MissForest techniques) when evaluated using different performance indicators. Moreover, the results show a trade-off between imputation accuracy and computational complexity with benchmark techniques showing a low computational complexity at the expense of accuracy when compared with our proposed technique.

Domain-adapted Gaussian mixture models for population-based structural health monitoring

Transfer learning, in the form of domain adaptation, seeks to overcome challenges associated with a lack of available health-state data for a structure, which severely limits the effectiveness of conventional machine learning approaches to structural health monitoring (SHM). These technologies utilise labelled information across a population of structures (and physics-based models), such that inferences are improved, either for the complete population, or for particular target structures — enabling a population-based view of SHM. The aim of these methods is to infer a mapping between each member of the population’s feature space (called a domain) in which a classifier trained on one member of the population will generalise to the remaining structures. This paper introduces the domain-adapted Gaussian mixture model (DA-GMM) for population-based SHM (PBSHM) scenarios. The DA-GMM, infers a linear mapping that transforms target data from one structure onto a Gaussian mixture model that has been inferred from source data (from another structure). The proposed model is solved via an expectation maximisation technique. The method is demonstrated on three case studies: an artificial dataset demonstrating the approach’s effectiveness when the target domain differs by two-dimensional rotations; a population of two numerical shear-building structures; and a heterogeneous population of two bridges, the Z24 and KW51 bridges. In each case study, the method is shown to provide informative results, outperforming other conventional forms of GMM (where no target labelled data are assumed available), and provide mappings that allow the effective exchange of labelled information from source to target datasets.