Structural health monitoring (SHM) plays a pivotal role in ensuring the safety, reliability and service life of engineering structures. In smart structures, networks of active-response materials (e.g. piezoelectric films, magnetostrictive patches, or fibre-optic cables), which convert mechanical and thermal stimuli into electrical or optical signals and act as the primary interface for continuous condition assessment. A persistent challenge is the influence of environmental and operational variability (EOV), particularly temperature changes, which can distort sensor measurements and either obscure or mimic genuine indicators of structural damage. Although numerous methodologies have been proposed to address this issue across various sensing platforms, a comprehensive comparative assessment across methodological categories remains lacking. This review critically examines 3 principal approaches developed to mitigate EOV: direct baseline compensation, adaptive and multi-baseline strategies, and reference-free techniques, including recent advances in transfer learning and hybrid physics-informed machine learning frameworks. A structured literature search spanning Scopus, Web of Science, IEEE Xplore and ScienceDirect underpins the analysis. Each approach is systematically evaluated, highlighting key benefits, limitations and suitability for varying operational scenarios. In addition, emerging trends, gaps and future research directions are identified, emphasising the need for hybrid models, real-time reference-free methodologies, robust uncertainty quantification and scalable population-based SHM solutions. The synthesis is intended to inform the design of next-generation smart, adaptable SHM systems regardless of sensing modality and their seamless integration into intelligent structures operating under complex real-world environmental conditions.

Ultrasonic guided waves (UGW) are widely used in structural health monitoring (SHM) systems due to the sensitivity of their propagation mechanisms to local material changes, i.e., those induced by damage. Post-processing of the signals gathered by piezoelectric sensors, typically used for both the excitation and the sensing of UGW, is a fundamental step to extract all the peculiar features that can be related to both damage location and severity. This research probes the efficacy of machine learning (ML) models in discerning damage location (R-Classification) and size (S-Classification). Seven supervised ML classifiers were examined: Ensemble-Subspace K-Nearest Neighbors (KNN), Ensemble-Bagged Trees, KNN-Fine, Ensemble-Boosted Trees, Support Vector Machine (SVM), Linear Discriminant, and SVM-Quadratic. The experimental dataset comprised measurements from varied reversible damage configurations on a composite panel, represented by wooden cuboids of single and three different sizes. Signal noise was minimized by performing a low-pass filter, and sequence forward selection-aided feature selection. The optimized ensemble classifier proved to be the most precise for R-Classification (95.83% accuracy), while Ensemble-Subspace KNN excelled in S-Classification (98.1% accuracy). This method offers accurate, efficient damage diagnosis and classification in composite structures, promising potential applications in aerospace, automotive, and civil engineering sectors.

In the condition monitoring of rotating systems, overfitting is a common challenge due to limited data history, which reduces the effectiveness of fault detection frameworks; this limitation often leads to unreliable diagnostics, resulting in unexpected machine failures and increased operational costs in industrial applications. Advances in deep learning suggest using simulated data to address this issue, but operational variabilities still cause significant data distribution shifts, affecting model accuracy. This article presents a new vibration-based monitoring framework that improves fault detection in rotary machines by effectively managing these shifts. It features a novel fine-tuning approach within sequential domain adaptation, requiring only a limited number of observations from the target domain for accurate model adjustment. The domain adaptation process is elucidated through a novel visualization of internal activation patterns within the sequential network. This method is further enhanced by a hybrid algorithm that combines wavelet transformation, a multi-layer perceptron, and a transformer encoder, followed by domain-specific fine-tuning. The framework’s effectiveness is demonstrated through experimental data from two different rotor systems, validated by sensitivity and comparative analyses, highlighting its robustness, generalizability, and practical applicability as a baseline in industrial fault detection scenarios.

This article compares two intelligent methods for automatic detection of unbalancing, cracks, and parallel misalignment in rotary machines. The finite element method is used to model the faults in a rotating system. The modeled system then operates virtually under different conditions in the steady-state operation; the vibrational responses are calculated numerically. To compare the accuracy of different manners in the classification of defective systems, firstly, four distinct types of features, i.e., statistical, frequency, time–frequency, and uncertainty are exploited. The T test process is utilized to test the extracted characteristics; the unreliable features are removed from feature vectors, then the remained ones are used in four supervised machine learning classifiers, i.e., support vector machine, k-nearest neighbors, Naive Bayes, and decision trees. In the following, as the convolution neural networks (CNNs) approach, the persistence spectrums of raw signals are plotted, and these graphs are introduced as input data. Comparing results of the different classification methods, it has been observed that although CNNs based on persistence spectrum graphs are computationally heavy and time-consuming, they provide more accurate results than the other classifiers. The results show that the proposed approach for rotor fault detection is effective, accurate, and robust and that it has promise for real engineering applications.

The goal of this research study is to classify damages in an aluminum panel using ultrasonic-guided waves (UGWs). The study is carried out on the premise that only a limited dataset, consisting of only two receivers, is available. In the experimental phase, three distinct parts, that is, stainless steel masses, were located on the surface of the panel, assumed to be mapped into four triangular regions, to experimentally simulate the effects induced by circular artificial damage of varied sizes. UGW-driven damage classification was thus performed via machine learning. A data synthesis technique called conditional generative adversarial network (cGAN) was used to generate 2000 samples (signals) for each of the 12 different malfunction scenarios (classes), based on the combination of damage region and size. Damage features were extracted using the wavelet time scattering technique, while the classification task was performed via a long short-term memory network with tuned hyperparameters. A validation procedure was executed to ensure that the classification network was not overfitted. Furthermore, two testing stages were performed to evaluate the designed framework. The designed framework was thus tested on the observations excluded from the training and validation phases, as well as on 10 additional novel experimental samples. The high accuracy and F1-score values of the validation and the two testing stages—97.7%, 98.5%, and 96.6%, respectively—attest to the generality of this methodology, which is neither underfitted nor overfitted. To further assess the robustness of the methodology, the cGAN was tested on a new sensor placement, demonstrating a consistent performance (95.3%) and confirming the framework applicability under different sensor configurations, showcasing potential extension to other classification problems featuring a limited dataset.

A growing interest in intelligent fault detection may sometimes lead to practical issues when existing malfunctions reveal analogous indications and the number of observations is limited. This article addresses the classification problem of two identical malfunctions, i.e., unbalancing and shaft bow in rotary machines, where only 56 observations were utilized for the training. The faulty systems are modeled in ABAQUS/CAE; a data set for each fault is created by simulation under various physical and operational conditions employing the uncertainty concept. The wavelet time scattering (WTS) technique extracts low-variance presentations from signals. With respect to the classification procedure of the faulted rotor systems, two models are examined with the extracted features from WTS as the input. Initially, a long short-term memory (LSTM) network is trained and tested, and then, the capability of a support vector machine (SVM) model is inquired. Ultimately, the classification models are trained and tested using the raw time series data and the extracted features to compare the effectiveness of the suggested methods, i.e., WTS. The employed approach for feature extraction demonstrated remarkable effectiveness in addressing a potential hurdle in identifying faults in rotating systems: the ability to differentiate between unbalanced and bowed rotors, irrespective of the classification model utilized.