Machine condition monitoring increasingly depends on distributed sensing, edge intelligence, and cloud analytics, yet timely and trustworthy health assessment remains constrained by latency, bandwidth, privacy, and reliability requirements. Cloud-only architectures provide scalable computation and historical data integration but often fail to satisfy real-time industrial needs, whereas edge-only deployments are limited by restricted computing resources and fragmented local knowledge. Edge–cloud collaboration has, therefore, emerged as a practical architecture for distributing perception, inference, learning, and coordination across hierarchical industrial systems. This review examines 147 publications on edge–cloud collaboration for machine condition monitoring published between 2019 and February 2026. A four-dimensional taxonomy is developed to organize the literature into model-centric, data-centric, resource and task-centric, and architecture and trust-centric mechanisms, while 13 survey and review papers are considered separately for contextual comparison. On this basis, the review analyzes representative collaboration mechanisms and enabling technologies, with particular attention to federated learning, transfer learning, knowledge distillation, digital twins, and deep reinforcement learning, and surveys their deployment in manufacturing, energy, transportation, and infrastructure monitoring scenarios. The literature remains dominated by model-centric collaboration, while architecture and trust-centric studies increasingly provide the system foundations required for practical deployment. The review further identifies major open challenges, including robust generalization under changing operating conditions, efficient data transmission, real-time resource coordination, interoperability, and trustworthy large-scale deployment, and outlines future directions in foundation-model-based edge–cloud collaboration, continual learning, dual digital twins, trustworthy collaboration, and privacy-preserving industrial ecosystems.

Machine condition monitoring using acoustic signals is essential for ensuring the reliability, safety, and efficiency of industrial machinery. Traditionally, fault detection relied on manual inspections, vibration sensors, and rule-based systems. These conventional methods were limited in capturing subtle audio variations produced by machines such as engines, compressors, and motors. They often required expert interpretation and were prone to errors, particularly in noisy industrial environments. Machine learning models based on handcrafted features, such as Mel-Frequency Cepstral Coefficients (MFCC) with Logistic Regression (LRC), Linear Discriminant Analysis (LDA), offered some automation but struggled with generalization, misclassifying acoustically similar faults like air leaks, idling disturbances, and oil leak variations. To address these limitations, this research proposes a hybrid intelligent framework, Hidden Unit Tree (HUT), which combines Hidden-Unit Bidirectional Encoder Representations from Transformers (HuBERT)-based deep audio embeddings with a Tree Alternating Optimization (TAO) Tree classifier. HuBERT, a transformer-based self-supervised model, captures high-level, contextual acoustic features, while the TAO Tree classifier provides robust non-linear decision-making. The system is implemented in a Tkinter graphical user interface (GUI), enabling end-to-end functionality including dataset upload, MFCC and HuBERT feature extraction, model training, evaluation, and real-time audio fault prediction. This research successfully integrates deep learning and advanced classification techniques to deliver a reliable, automated machine condition monitoring solution. The combination of rich acoustic feature extraction and robust classification ensures accurate fault detection, making the system suitable for predictive maintenance and Industry 4.0 applications, thereby completing a fully functional, real-time intelligent fault diagnosis tool.

Monitoring the condition of industrial machines through acoustic signals plays a critical role in maintaining operational reliability, safety, and efficiency. Conventional fault detection approaches, including manual inspections, vibration-based sensing, and rule-driven systems, often fall short in identifying subtle variations in machine-generated sounds from equipment such as motors, compressors, and engines. These methods typically depend on expert analysis and are susceptible to inaccuracies, especially in noisy environments. Earlier machine learning techniques that relied on handcrafted features such as Mel-Frequency Cepstral Coefficients (MFCC) combined with classifiers like Logistic Regression and Linear Discriminant Analysis introduced partial automation but faced challenges in generalization. In particular, they struggled to distinguish between acoustically similar fault types, including air leaks, idle irregularities, and oil-related anomalies. To overcome these limitations, this study presents a hybrid intelligent approach termed the Hidden Unit Tree (HUT). This framework integrates deep audio representations derived from Hidden-Unit Bidirectional Encoder Representations from Transformers (HuBERT) with a Tree Alternating Optimization (TAO) Tree classifier. The HuBERT model, built on transformer architecture and trained in a self-supervised manner, effectively captures complex and contextual acoustic patterns. Meanwhile, the TAO Tree enhances classification performance through flexible and non-linear decision boundaries. The proposed system is implemented within a Tkinter-based graphical user interface, offering a complete pipeline that includes data uploading, feature extraction using both MFCC and HuBERT, model training, performance evaluation, and real-time fault prediction from audio inputs. By combining advanced feature learning with a robust classification strategy, this work delivers an efficient and automated solution for machine condition monitoring. The resulting system achieves reliable fault detection, making it well-suited for predictive maintenance and Industry 4.0 environments as a comprehensive, real-time intelligent diagnostic tool.

Acoustic-based monitoring of machine health is a vital approach for improving the dependability, safety, and performance of industrial systems. Earlier fault identification techniques mainly depended on manual inspection, vibration analysis, and predefined rule-based mechanisms, which were often insufficient for detecting fine-grained acoustic changes produced by equipment such as motors, engines, and compressors. These approaches required significant human expertise and were highly sensitive to environmental noise, leading to inconsistent results. Even with the introduction of machine learning models using engineered features like MFCC combined with LRC and LDA, the performance remained limited due to poor adaptability and difficulty in differentiating closely related fault conditions such as air leaks, oil leak variations, and idle disturbances. To address these shortcomings, the proposed system adopts a hybrid architecture called Hidden Unit Tree (HUT), leveraging HuBERT for extracting deep, context-aware audio embeddings and a TAO Tree classifier for effective non-linear classification. HuBERT enhances feature representation by learning complex acoustic patterns through selfsupervised training, while the TAO Tree improves decision accuracy through optimized tree-based learning. The entire framework is deployed within a Tkinter-based GUI that supports complete workflow execution, including dataset handling, MFCC and HuBERT feature extraction, model training, evaluation, and live fault prediction. This integrated approach significantly improves fault recognition capability, offering a robust and intelligent solution for real-time machine condition monitoring aligned with predictive maintenance and Industry 4.0 requirements.

Optical transduction-enabled acoustic sensing stands out as a promising substitute for conventional capacitive sensing in MEMS microphones, especially for ultra-low-noise applications. Here, we demonstrate a prototype optical MEMS microphone sensor with optimized photonic-mechanical coupling. The sensor is fabricated through a standard micro-manufacturing process, wherein the diaphragm is made of polysilicon material with low bending stiffness and compatibility with the MEMS process. The sensor enables measurement within a frequency bandwidth of 200 Hz to 6.1 kHz. A sensitivity of 89.2 mV/Pa is obtained within the sound pressure measurement range of 0.1 to 1 Pa with a maximum nonlinearity error of 2.97% and a repetitive error of 1.11%. The signal-to-noise ratio of the sensor is 70.2 dB based on the balanced differential method. Moreover, the total harmonic distortion of the sensor is 0.08% within the pressure range. The sensor demonstrates significant application potential in high-performance acoustic detection, such as medical auscultation and machine condition monitoring.

This dataset provides monitoring data from a rotating electromechanical system under controlled and faulty conditions. The system is equipped with heterogeneous sensors, including accelerometers, current and temperature to capture its physical behavior across a range of stationary and non-stationary rotation's speeds. Different faults, and some of them even with different severities, were systematically introduced in components of the electromechanical drivetrain-such as misalignments, bearing defects, and unbalances-to simulate degradation scenarios typically encountered in industrial settings. The resulting multivariate time series data are suitable for a variety of applications, including machine learning-based diagnostics, signal processing, and condition monitoring. The availability of multiple sensor modalities enables advanced techniques such as information fusion and multi-sensor data analysis. The experiments include variable speed conditions, introducing dynamic complexities that enhance the dataset's realism and usefulness for robust algorithm development. This paper describes the experimental setup, sensor placement, fault injection procedures, and data acquisition parameters in detail.

Numerical simulation-based digital twins are emerging as a transformative technology capable of significantly enhancing operational efficiency and minimising costly maintenance and human intervention for smart products and services. However, the inherent limitations of physical monitoring and the uncertainties associated with product or service performance can be effectively addressed through the strategic application of parameterised numerical models combined with advanced machine learning (ML) algorithms. To address the research gap, this research investigates how a novel and systematic digital-twin-based design and analysis approach can facilitate the transformation of a conventional Shell-and-Tube Heat Exchanger (STHE) into a smart machine within the evolving framework of Industry 4.0. The methodology involves devising a data-driven digital twin (DT) for the STHE, utilising coupled Computational Fluid Dynamics (CFD) and Finite Element Analysis (FEA) simulations to numerically investigate key performance parameters. This process enables the integration of virtual sensors, the fusion of physical measurements, and the deployment of advanced machine learning algorithms to precisely identify critical points for sensor and actuator placement in an Internet of Things (IoT)-based machine condition monitoring (MCM) system. The developed STHE digital twin successfully demonstrates its capability to extract data of key performance parameters, which enables seamless integration with sensors and actuators. This digital twinning further empowers a digital design and prototyping process for visualiszed, real-time IoT based machine condition monitoring. Advanced ML algorithms are employed to identify the locations of the critical points in the STHE, where sensors are installed for IoT-based MCM. The developed STHE DT demonstrates its capability of extracting crucial data based on fundamental principles of mass, energy, and momentum, facilitating seamless integration with sensors. The underpinning concept, comprehensive methodological framework, and practical implementation process of the STHE digital twin presented herein provide a robust foundation. This work represents a significant scientific contribution towards enabling the transformation of conventional mechanical systems into intelligent, data-driven smart products, aligning with the objectives of Industry 4.0.

Posterior probability perspective for obtaining physically interpretable optimized weights for machine condition monitoring

Abstract This study introduces VibroScan, a low-cost, open-source vibration monitoring system developed using microelectromechanical systems-based accelerometers and ESP32 microcontrollers, tailored for educational and field applications. The system enables real-time triaxial acceleration data acquisition, frequency-domain anomaly detection via fast Fourier transform and machine learning algorithms, and remote monitoring through a message queuing telemetry transport-based web interface. Designed and tested on a metal casting machine, the system successfully identified dominant spectral behaviors and operational anomalies during casting cycles. As part of the hands-on engineering experience, students contributed to the design of the sensor enclosure, system logo, and a responsive web dashboard built with HTML, JavaScript, and Chart.js. These tasks enhanced their understanding of not only embedded systems and signal processing, but also product development and system integration. Comparative analysis of Z score and long-short term memory (LSTM)-based models revealed that while both methods detect spectral anomalies effectively, LSTM demonstrated superior sensitivity to dynamic frequency shifts. Beyond industrial diagnostics, this study is also of value in seismology with for example implications of machine health monitoring in earthquake early warning contexts. Many industrial machines suffer significant economic losses during seismic events due to undetected resonance or structural failure. Deploying smart sensors like VibroScan in critical facilities enables dual function monitoring continuous machinery assessment and preseismic anomaly tracking offering a practical step toward infrastructure resilience. VibroScan exemplifies how accessible hardware and modern signal analytics can be leveraged in Science, Technology, Engineering, and Mathematics curricula for students of engineering and seismology to bridge theoretical learning and real world problem solving. It provides students with direct exposure to data-driven maintenance systems, Internet of Things architectures, and the societal impact of smart instrumentation key competencies for future engineers and researchers in seismology, structural monitoring, and beyond.

This paper presents a Proof of Concept (PoC) for PredictiveMaintenance MCP, an open-source server based on the Model Context Protocol (MCP) that supports machine condition monitoring and predictive maintenance via natural language interaction with Large Language Models (LLMs). The server constrains the LLM within an explicit perimeter of deterministic resources and tools for vibration-based diagnostics, including FFT spectral analysis with peak identification, envelope analysis for rolling element bearing defects, time-domain indicators, vibration severity assessment consistent with ISO standards and semi-supervised anomaly detection on extracted features. Each tool invocation produces structured outputs and artifacts that record inputs, parameters, and results. The LLM acts as an orchestrator that selects resources, configures parameters, invokes tools, and synthesizes conclusions anchored to computed evidence, thereby improving traceability and repeatability compared to unconstrained text-only interaction. End-to-end workflows are demonstrated in a reproducible package with code, examples, and demo data to support community-driven validation and extension toward industrial requirements. The software is archived on Zenodo and the GitHub repository serves as the collaboration hub.

Condition monitoring and fault diagnosis of synchronous machines–A review

Artificial intelligence (AI) is increasingly adopted in manufacturing for tasks such as automated inspection, predictive maintenance, and condition monitoring. However, the opaque, black-box nature of many AI models remains a major barrier to industrial trust, acceptance, and regulatory compliance. This study investigates how explainable artificial intelligence (XAI) techniques can be used to systematically open and interpret the internal reasoning of AI systems commonly deployed in manufacturing, rather than to optimise or compare model performance. A unified explainability-centred framework is proposed and applied across three representative manufacturing use cases encompassing heterogeneous data modalities and learning paradigms: vision-based classification of casting defects, vision-based localisation of metal surface defects, and unsupervised acoustic anomaly detection for machine condition monitoring. Diverse models are intentionally employed as representative black-box decision-makers to evaluate whether XAI methods can provide consistent, physically meaningful explanations independent of model architecture, task formulation, or supervision strategy. A range of established XAI techniques, including Grad-CAM, Integrated Gradients, Saliency Maps, Occlusion Sensitivity, and SHAP, are applied to expose model attention, feature relevance, and decision drivers across visual and acoustic domains. The results demonstrate that XAI enables alignment between model behaviour and physically interpretable defect and fault mechanisms, supporting transparent, auditable, and human-interpretable decision-making. By positioning explainability as a core operational requirement rather than a post hoc visual aid, this work contributes a cross-modal framework for trustworthy AI in manufacturing, aligned with Industry 5.0 principles, human-in-the-loop oversight, and emerging expectations for transparent and accountable industrial AI systems.

Revisiting Lempel-Ziv complexity in machine condition monitoring: a fresh perspective

MSANet: Multi-Stage attention network for anomalous sound detection in machine condition monitoring

Non-iterative optimal blind deconvolution and its application to machine condition monitoring

Background: The increasing complexity of industrial production systems requires machine condition monitoring solutions that are capable of operating in real time with high accuracy and responsiveness to support predictive maintenance strategies. Conventional cloud based monitoring systems often experience limitations such as high latency and dependence on stable network connectivity, which can delay decision making processes in critical industrial operations. Objective: This study aims to design and evaluate an Industrial Internet of Things (IIoT) architecture based on edge computing to improve the efficiency of industrial sensor data processing and accelerate anomaly detection in industrial machines. Method: The research adopts an experimental approach by designing a system architecture consisting of a sensor layer, edge computing layer, and cloud layer. Industrial sensors, including vibration, temperature, and current sensors, continuously collect machine operational data, which are then processed locally at the edge node using a machine learning based anomaly detection algorithm. System testing is conducted in a simulated manufacturing environment to evaluate performance based on latency, reliability, and detection accuracy. Results: The results indicate that edge based data processing significantly reduces latency compared with cloud-based processing and enables faster responses to machine condition changes. Additionally, the implemented anomaly detection algorithm achieves high accuracy in identifying abnormal sensor data patterns.

ABSTRACT In intelligent manufacturing systems, the machine condition and the product quality data are conveniently available. Developing an effective maintenance strategy based on these two sources can significantly enhance cost-effectiveness and system reliability. However, existing studies have primarily focused on maintenance strategies based solely on either machine condition or product quality monitoring, while their joint monitoring has received limited attention. This paper proposes a novel strategy for monitoring manufacturing machines by integrating degradation and quality data, and designs a fixed-number variable inspection interval (VII) scheme to address the high cost of frequent sampling. First, the machine degradation process is modelled by a Wiener process, and the relationship between machine degradation and product quality is established. The variation of product quality is monitored using the S 2 control chart, and a fixed-number VII scheme is designed. Second, a joint condition-based maintenance (JCBM) strategy is formulated, including the derivation of the expected time, cost, and availability for each maintenance scenario. Optimal maintenance plans are determined by minimizing the expected cost rate (ECR). Finally, a case study on scroll production in an air compressor system demonstrates the proposed strategy, showing that JCBM outperforms traditional strategies in both cost-effectiveness and monitoring performance.

In additive manufacturing, mechanical vibrations generated during the printing process produce characteristic acoustic emissions, which are directly influenced by toolpath kinematics. These vibrations can adversely affect dimensional accuracy and interlayer adhesion, underscoring the need for effective process monitoring. This study investigates the correlation between specific toolpath geometries and their acoustic signatures in a Fused Deposition Modeling (FDM) 3D printer to establish a foundation for non-invasive condition monitoring. Five fundamental motion patterns—diagonal (Quadrants I-III and II-IV), horizontal, cylindrical, and vertical—were fabricated in an anechoic chamber. Acoustic emissions were captured via two microphones positioned 5 cm from the printer and analyzed in the time domain using statistical features: Root Mean Square (RMS), Kurtosis, and Crest Factor. The measured sound pressure levels ranged from 0.5 Pa to 1.5 Pa. Results indicate that the vertical toolpath yielded the lowest RMS (0.0863) and Crest Factor (5.38) values, reflecting the least intense and most stable acoustic emission. Conversely, diagonal patterns exhibited significantly higher values, denoting greater vibrational energy and transient fluctuations. These findings demonstrate a definitive influence of motion geometry on a printer's acoustic signature. The vertical pattern is identified as the most stable under the tested parameters. This research confirms that time-domain acoustic analysis is a viable technique for characterizing machine performance. Establishing this baseline correlation enables the future development of real-time, sound-based monitoring systems capable of predicting print defects and facilitating predictive maintenance, thereby enhancing the reliability and quality of additive manufacturing processes.

Data is an important resource for gaining knowledge about the behavior and condition monitoring of machines, enabling the estimation of parameters and the prediction of failures. However, in industrial environments, sensor interruptions often create gaps in the time series, which affects the reliability of the data. To overcome this challenge, this paper proposes an imputation strategy based on recurrent neural networks, in particular long short-term memory (LSTM) models, within a multivariate encoder–decoder architecture. This approach utilizes correlations between variables to reconstruct missing values, resulting in more complete and robust datasets. Experimental results with multivariate time series show that the proposed method achieves accurate imputation, with errors as low as RMSE = 2.33 and R2 = 0.90 for some variables. Comparisons with alternative architectures, including GRU and Dense networks, show that LSTM excels in specific cases (e.g., VL3, R2 = 0.45), while the Dense architecture provides more stable performance across most variables. In particular, the Dense model achieved the best overall balance between accuracy and robustness, reaching RMSE = 2.33 and R2 = 0.90 for the best-performing variables, while the LSTM achieved the lowest error values in targeted scenarios, confirming its suitability for capturing complex temporal dependencies. Overall, this study highlights the feasibility of using recurrent neural networks to exploit temporal correlations for reliable data recovery, even under conditions of signal interruption in factory environments.

Abstract Vibration monitoring for ball screw feed systems has always been a prominent research focus globally. The event camera, as an emerging bio-inspired sensor technology, offers significant advantages including non-contact operation, ultra-low latency, and high dynamic range, leading to its increasing adoption in machine condition monitoring in recent years. In view of the limitations of traditional vibration sensors—such as single-point measurement and complex installation—this study applies event cameras to identify abnormal vibrations in ball screw feed systems. To address the inadequacy of existing event camera-based vibration monitoring methods for ball screw feed systems, an Attention Mechanism Enhanced Hybrid Neural Network (AMHNet) is proposed. Leveraging the inherent temporal, channel, and spatial characteristics of event camera data, specialized attention mechanisms augment feature extraction capabilities across these three dimensions. This architecture significantly improves identification performance across varying vibration conditions. An experimental platform was constructed, and a tailored event-based vision dataset was generated for abnormal vibration identification. Experimental results show that the proposed AMHNet achieves high identification accuracy, confirming the method’s feasibility and offering a novel approach for vibration monitoring in ball screw feed systems.