Vibration-Based Condition Monitoring of Diesel Engines in Industrial Energy Applications: A Scoping Review

Abstract

Diesel engines remain the foundation for obtaining mechanical energy in sectors where autonomy and reliability are required; however, predictive diagnostics under real-world conditions remain challenging. The purpose of this scoping review is the investigation and systematization of published scientific data on the application of vibration methods for monitoring the technical condition of diesel engines in industrial or controlled laboratory conditions. Based on numerous results of publication analysis, sensor configurations, diagnosed components, signal analysis methods, and their application for assessing engine technical condition are considered. As methods for determining vibration parameters, time-domain and frequency-domain analysis, adaptive decompositions, and machine and deep learning algorithms predominate; high accuracy is more often achieved under controlled conditions, while confirmations of robustness on industrial installations are still insufficient. Key limitations for the application of vibration monitoring methods include the multicomponent and non-stationary nature of signals, a high level of noise, requirements for sensor placement, communication channel limitations, and the need for on-site processing; meanwhile, the assessment of torsional vibrations remains technically challenging. It is concluded that field validations of vibroacoustic data, the use of multimodal sensor platforms, noise-immune algorithms, and model adaptation to the specific environment are necessary, taking into account fuel quality, transient conditions, and climatic factors.

1. Introduction

Diesel engines are a widely used source of mechanical energy in many industries, converting fuel energy into mechanical work for drives of ships, vehicles, electric generators, and other consumers [1]. Furthermore, equipment with diesel engines is used in all types of mining enterprises [2]; in underground conditions, diesel monorail transport is widely used [3,4]. Diesel engines are chosen for their high thermal efficiency, operational reliability, and fuel flexibility, especially under high loads.

However, prolonged operation under harsh and non-stationary conditions accelerates their mechanical wear, leading to an increased risk of critical failures and unplanned downtime. For diesel engines, a close connection is characteristic between vibration anomalies and the emergence of specific defects (piston wear, crankshaft imbalance, etc.) [5], which is why, under these conditions, vibration condition monitoring has established itself as a method of non-destructive testing for rotational and reciprocating components of diesel engines.

Continuous monitoring of engine vibrations allows for the detection of early signs of their malfunctions and the prevention of failures, reduces maintenance costs, and maintains specified reliability indicators [6,7]. For the mineral resource sector, vibration monitoring of engines is particularly promising due to non-stationary regimes and harsh operating conditions, as the method is resistant to regime variability [8] and is suitable for monitoring the condition of heavy equipment (bulldozers, excavators) [9,10].

At the same time, vibration monitoring has a number of fundamental characteristics. Vibration signals are multicomponent and non-stationary; in real operation, they are superimposed with environmental disturbances and background vibrations from coupled equipment. This imposes high demands on measuring instruments, their configurations, signal processing robustness, and algorithms functioning in real-time.

Currently, new approaches are being developed in various areas for solving vibration monitoring tasks [11,12]: intelligent recognition methods based on the application of swarm intelligence algorithms [13]; robust procedures for decomposing vibration signals, reducing mode mixing and edge distortions during signal decomposition [14]; fusion of heterogeneous data based on the joint analysis of vibrations and wear debris, providing an accuracy increase for early detection of about 16% compared to the isolated application of methods [15]; integrated vibroacoustic schemes and predictive maintenance loops for engines, focused on reducing fuel consumption, emissions, and increasing service life [16].

An analysis of the results from 77 studies (see

Table A1. Overview of reviewed studies.

Title	Ref.	Application Aspect	Diagnostic and Functional Targets	Digital and Modeling Technologies
Combustion Noise Analysis for Combustion and Fuels Diagnosis of a Compression Ignition Diesel Engine Operating with Biodiesels	[20]	Transportation	Combustion Dynamics, Cylinders	Classical Signal Processing
Combustion parameter estimation for ICE from surface vibration using frequency spectrum analysis	[21]	Power Generation	Combustion Dynamics	Hybrid/Integrated Architectures (Classical Signal Processing + Numerical Simulations)
Coaxial Flexible Fiber-Shaped Triboelectric Nanogenerator Assisted by Deep Learning for Self-Powered Vibration Monitoring	[22]	General Industrial Installations	Vibration Transmission Paths	Deep Learning
Combination of Optimized Variational Mode Decomposition and Deep Transfer Learning: A Better Fault Diagnosis Approach for Diesel Engines	[23]	General Industrial Installations	Rotating Assemblies	Hybrid (Deep Learning + Classical Signal Processing)
Diesel Engine Gearbox Fault Diagnosis Based on Multi-features Extracted from Vibration Signals	[24]	General Industrial Installations	Rotating Assemblies	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing)
Diesel Engine Turbocharger Monitoring by Processing Accelerometric Signals through Empirical Mode Decomposition and Independent Component Analysis	[25]	Transportation	Turbocharging	Hybrid/Integrated Architectures (Classical Signal Processing + Other)
Turbocharger speed estimation via vibration measurements for combustion sensing	[26]	Transportation	Turbocharging	Classical Signal Processing
On the use of cyclostationary indicators in IC engine quality control by cold tests	[27]	Transportation	Cylinders, Rotating Assemblies	Classical Signal Processing
Diagnosing Cracks in the Injector Nozzles of Marine Internal Combustion Engines during Operation Using Vibration Symptoms	[28]	Marine and Ship Propulsion Systems	Fuel Injection	Machine Learning
Spectral Analysis of Torsional Vibrations Measured by Optical Sensors, as a Method for Diagnosing Injector Nozzle Coking in Marine Diesel Engines	[29]	Marine and Ship Propulsion Systems	Fuel Injection	Classical Signal Processing
Vibroactivity analysis of a dual fuel diesel engine based on the knock sensor signal and measuring pressure in the combustion chamber	[30]	Transportation	Combustion Dynamics	Classical Signal Processing
Classification-Based Fuel Injection Fault Detection of a Trainset Diesel Engine Using Vibration Signature Analysis	[31]	Transportation	Fuel Injection, Combustion Dynamics	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing + Dimensionality Reduction)
Engine knock detection for a multifuel engine using engine block vibration with statistical approach	[32]	Transportation	Combustion Dynamics	Statistical Analysis and Threshold-Based Alarms
Analysis of transmission pathways of combustion-induced vibration in a diesel engine using wavelet cross-correlation analysis method	[33]	Transportation	Combustion Dynamics, Vibration Transmission Paths	Classical Signal Processing
Investigation of combustion-induced vibration sources in a diesel engine in the time-frequency domain using the wavelet analysis method and wavelet cross-correlation analysis method	[34]	Transportation	Combustion Dynamics, Pistons, Crankshaft Torsional Behavior Speed Fluctuations, Vibration Transmission Paths	Classical Signal Processing
Traction diesel engine anomaly detection using vibration analysis in octave bands	[35]	Transportation	Rotating Assemblies	Machine Learning
Analysis of Vibration on an Engine Block Caused by Combustion in a Diesel Engine	[36]	Transportation	Crankshaft Torsional Behavior, Pistons	Classical Signal Processing
Torsional Vibration Stress and Fatigue Strength Analysis of Marine Propulsion Shafting System Based on Engine Operation Patterns	[37]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations	Numerical Simulations
Torsional system dynamics of low speed diesel engines based on instantaneous torque: Application to engine diagnosis	[38]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations, Combustion Dynamics, Cylinders	Hybrid/Integrated Architectures (Numerical Simulations + Machine Learning)
Research on the Field Dynamic Balance Technologies for Large Diesel Engine Crankshaft System	[39]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations	Classical Signal Processing
Assessing combustion performance of a diesel reciprocating engine under various fuel blends using a calculus-statistical time-series vibration based approach	[40]	Power Generation	Combustion Dynamics	Statistical Analysis and Threshold-Based Alarms
Refined composite multiscale fuzzy entropy based fault diagnosis of diesel engine	[41]	General Industrial Installations	Valves, Air Management Systems, Cylinders	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning)
Vibration and acoustic characteristics of a city-car engine fueled with biodiesel blends	[42]	Transportation	Combustion Dynamics	Classical Signal Processing
Methods of Real-Time Parametric Diagnostics for Marine Diesel Engines	[43]	Marine and Ship Propulsion Systems	Fuel Injection, Valves, Combustion Dynamics, Cylinders, Pistons	Hybrid/Integrated Architectures (Edge Monitoring + Classical Signal Processing)
Optimization of Transducer Location for Novel Non-Intrusive Methodologies of Diagnosis in Diesel Engines	[44]	Transportation	Combustion Dynamics	Classical Signal Processing
Assessment of in-cylinder pressure in diesel engines using novel combustion indices	[45]	Transportation	Combustion Dynamics	Statistical Analysis and Threshold-Based Alarms
Combustion parameter evaluation of diesel engine via vibration acceleration signal	[46]	Transportation	Combustion Dynamics	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing + Statistical Compression)
Combustion parameters estimation based on multi-channel vibration acceleration signals	[47]	Transportation	Combustion Dynamics	Hybrid/Integrated Architectures (Classical Signal Processing + Statistical Analysis).
Research on Diesel Engine Fault Diagnosis Method Based on Stacked Sparse Autoencoder and Support Vector Machine	[48]	General Industrial Installations	Fuel Injection, Valves, Air Management Systems, Lubrication/Oil Supply System	Hybrid/Integrated Architectures (Machine Learning + Deep Learning)
Simulation and modal analysis of marine diesel engine based on finite element model and vibration sensor data	[49]	Marine and Ship Propulsion Systems	Structural Supports	Hybrid/Integrated Architectures (Numerical Simulations + Experimental Modal Analysis)
Diesel engine vibration monitoring based on a statistical model	[50]	Marine and Ship Propulsion Systems	Vibration Transmission Paths	Statistical Analysis and Threshold-Based Alarms
Online Identification and Verification of the Elastic Coupling Torsional Stiffness	[51]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations	Numerical Simulations
Strategies for Mitigating Runout Interference in Torsional Vibration Measurement of Diesel Engine Crankshafts	[52]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations	Classical Signal Processing
Open-Source Data Logger System for Real-Time Monitoring and Fault Detection in Bench Testing	[53]	General Industrial Installations	Bearings, Pistons, Valves, General Condition	Hybrid/Integrated Architectures (IoT + Wireless Sensors + Classical Signal Processing)
Smart temperature and vibration monitoring device for small-scale fishing vessel engines	[54]	Marine and Ship Propulsion Systems	Rotating Assemblies	Hybrid/Integrated Architectures (IoT + Deep Learning)
Combustion parameters identification and correction in diesel engine via vibration acceleration signal	[55]	Transportation	Combustion Dynamics	Analytical Monitoring Frameworks
Engine Working State Recognition Based on Optimized Variational Mode Decomposition and Expectation Maximization Algorithm	[56]	Transportation	Fuel Injection, Valves	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning + Optimization)
Diesel engine injection faults detection and classification utilizing unsupervised fuzzy clustering techniques	[57]	Transportation	Fuel Injection	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning)
Fault detection and diagnosis of a 12-cylinder trainset diesel engine based on vibration signature analysis and neural network	[58]	Transportation	Combustion Dynamics, Fuel Injection, Cylinders	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing)
A Vibration Analysis for the Evaluation of Fuel Rail Pressure and Mass Air Flow Sensors on a Diesel Engine: Strategies for Predictive Maintenance	[59]	General Industrial Installations	Fuel Injection, Air Management Systems, Combustion Dynamics	Classical Signal Processing
Diesel Engine Valve Clearance Fault Diagnosis Based on Improved Variational Mode Decomposition and Bispectrum	[60]	General Industrial Installations	Valves	Classical Signal Processing
Fault diagnosis of valve clearance in diesel engine based on BP neural network and support vector machine	[61]	General Industrial Installations	Valves, Cylinders	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning + Optimization)
A New Fault Diagnosis Method for a Diesel Engine Based on an Optimized Vibration Mel Frequency under Multiple Operation Conditions	[62]	Marine and Ship Propulsion Systems	Valves	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing)
Compression Reconstruction and Fault Diagnosis of Diesel Engine Vibration Signal Based on Optimizing Block Sparse Bayesian Learning	[63]	General Industrial Installations	Valves, Combustion Dynamics, General Condition	Hybrid/Integrated Architectures (Advanced DSP: Compressive Sensing + Dictionary Learning)
A Condition-Monitoring Approach for Diesel Engines Based on an Adaptive VMD and Sparse Representation Theory	[64]	Transportation	Valves, Fuel Injection, Combustion Dynamics	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing + Optimization)
The Piston Slap Force Reconstruction of Diesel Engine Using WOA-VMD and Deconvolution	[65]	Transportation	Pistons, Cylinders	Hybrid/Integrated Architectures (Advanced Signal Processing + Optimization)
A Hybrid Dynamic Principal Component Analysis Feature Extraction Method to Identify Piston Pin Wear for Binary Classifier Modeling	[66]	Transportation	Pistons	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing + Feature Engineering)
Effects of Piston Scuffing Fault on the Performance and Vibro-Acoustic Characteristics of a Diesel Engine: An Experimental Study	[67]	General Industrial Installations	Pistons, Cylinders	Classical Signal Processing
An Experimental Study of the Effects of Cylinder Lubricating Oils on the Vibration Characteristics of a Two-Stroke Low-Speed Marine Diesel Engine	[68]	Marine and Ship Propulsion Systems.	Cylinders	Classical Signal Processing
Tribological effect of piston ring pack on the crankshaft torsional vibration of diesel engine	[69]	General Industrial Installations	Pistons, Cylinders	Numerical Simulations
Diesel engine diagnosis based on entropy of vibration signals and machine learning techniques	[70]	Transportation	Combustion Dynamics, Bearings	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning)
A novel adaptive fault diagnosis algorithm for multi-machine equipment: application in bearing and diesel engine	[71]	General Industrial Installations	Bearings, General Condition	Hybrid/Integrated Architectures (Deep Learning + Classical Signal Processing + Optimization)
Variational mode decomposition denoising combined with the Euclidean distance for diesel engine vibration signal	[72]	Transportation	Bearings	Classical Signal Processing
Quantitative misalignment detection method for diesel engine based on the average of shaft vibration and shaft shape characteristics	[73]	Marine and Ship Propulsion Systems	Rotating Assemblies	Numerical Simulations
Application of empirical mode decomposition and extreme learning machine algorithms on prediction of the surface vibration signal	[75]	General Industrial Installations	Vibration Transmission Paths	Machine Learning
Multi-Factor Operating Condition Recognition Using 1D Convolutional Long Short-Term Network	[77]	General Industrial Installations	Rotating Assemblies, Cylinders, Combustion Dynamics	Deep Learning
Improved deep residual shrinkage network for a multi-cylinder heavy-duty engine fault detection with single channel surface vibration	[78]	General Industrial Installations	Cylinders, Valves, General Condition	Deep Learning
Research on Diesel Engine Fault Status Identification Method Based on Synchro Squeezing S-Transform and Vision Transformer	[79]	General Industrial Installations	General Condition	Hybrid/Integrated Architectures (Deep Learning + Classical Signal Processing)
A Novel Fault Diagnosis Method for Diesel Engine Based on MVMD and Band Energy	[81]	Transportation	Cylinders	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning)
A new swarm intelligence optimized multiclass multi-kernel relevant vector machine: An experimental analysis in failure diagnostics of diesel engines	[82]	General Industrial Installations	Valves, Combustion Dynamics, General Condition	Machine Learning
Random convolutional neural network structure: An intelligent health monitoring scheme for diesel engines	[83]	Marine and Ship Propulsion Systems	Rotating Assemblies, Bearings	Deep Learning
Data-driven identification of rotating machines using ARMA deterministic parameter evolution in the angle/time domain	[84]	General Industrial Installations	Rotating Assemblies, Combustion Dynamics, Vibration Transmission Paths	Hybrid/Integrated Architectures (Statistical Analysis + Classical Signal Processing)
Transfer Diagnosis Model of Internal Combustion Engine With Embedded Vibration Signal Impact Decomposition	[80]	General Industrial Installations	Combustion Dynamics, Valves, Pistons, Bearings, Crankshaft Torsional Behavior Speed Fluctuations	Hybrid/Integrated Architectures (Deep Learning + Transfer Learning + Domain Adaptation)
A feature extraction and visualization method for fault detection of marine diesel engines	[85]	Marine and Ship Propulsion Systems	General Condition	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing)
Fault feature extraction of diesel engine based on bispectrum image fractal dimension	[86]	General Industrial Installations	Valves	Classical Signal Processing
Fault Classification in Diesel Engines Based on Time-Domain Responses through Signal Processing and Convolutional Neural Network	[87]	Marine and Ship Propulsion Systems	Crankshaft Torsional Behavior Speed Fluctuations, Cylinders	Hybrid/Integrated Architectures (Classical Signal Processing + Deep Learning)
Diagnostic and measurement system for marine engines	[88]	Marine and Ship Propulsion Systems	Combustion Dynamics, Fuel Injection, Cylinders, Bearings	Classical Signal Processing
Development and Validation of a Vibration-Based Virtual Sensor for Real-Time Monitoring NOx Emissions of a Diesel Engine	[98]	Transportation	Combustion Dynamics	Hybrid/Integrated Architectures (Classical Signal Processing + Machine Learning)
Diesel Engine Fault Diagnosis Method Based on Optimized VMD and Improved CNN	[99]	General Industrial Installations	General Condition	Hybrid/Integrated Architectures (Deep Learning + Machine Learning + Classical Signal Processing)
Identification of power output of diesel engine by analysis of the vibration signal	[100]	Marine and Ship Propulsion Systems	General Condition	Machine Learning
Method for assessing the relationship between the characteristics of vibroactivity and the design parameters of a marine diesel	[101]	Marine and Ship Propulsion Systems	Pistons, Cylinders, General Condition	Analytical Monitoring Frameworks
Performance map measurement, Zero-Dimensional modelling & vibration analysis of a single cylinder diesel engine	[102]	Transportation	Cylinders, Combustion Dynamics	Hybrid/Integrated Architectures (Classical Signal Processing + Numerical Simulations)
Piston scuffing fault and its identification in an IC engine by vibration analysis	[103]	Transportation	Pistons, Cylinders	Classical Signal Processing
Prediction and Analysis of Ship Engine Vibration Signals Based on Prompted Language Models	[104]	Marine and Ship Propulsion Systems	General Condition	Large Language Models
Study on the relationship between combustion parameters and cylinder head vibration signal in time domain	[105]	Transportation	Combustion Dynamics	Hybrid/Integrated Architectures (Numerical Simulations + Classical Signal Processing)
Time-frequency Feature Extraction Method of the Multi-Source Shock Signal Based on Improved VMD and Bilateral Adaptive Laplace Wavelet	[106]	General Industrial Installations	Valves	Hybrid/Integrated Architectures (Machine Learning + Classical Signal Processing)
Vibration analysis and combustion parameter evaluation of CI engine based on Fourier Decomposition Method	[107]	General Industrial Installations	Combustion Dynamics	Classical Signal Processing

, Appendix A) selected using search and screening procedures (see Section 2) was conducted. Unlike similar reviews summarizing test results in general, or tests limited to laboratory settings, this work focuses on the limitations of vibration monitoring for diesel engines integrated into real-world production processes.

The article is structured as follows.

Section 2 presents the scoping review methodology: search strategy, inclusion/exclusion criteria, screening and data extraction process (based on the PRISMA-ScR guidelines).

Section 3 synthesizes and groups the results along five analytical directions:

(i)

research landscape,

(ii)

measurement tools and sensor configurations,

(iii)

diagnostic and functional objectives,

(iv)

digital technologies and models used in vibration monitoring,

(v)

limitations of using vibration monitoring for diesel engines.

Section 4 presents a discussion of the identified methodological foundations, conceptual models, and structural barriers.

Section 5 formulates the conclusions, highlights under-researched areas, and outlines directions for future work in developing vibration diagnostic tools for diesel engines operating in complex industrial environments.

The objective of this scoping review is to investigate and systematize published scientific data on the application of vibration methods for monitoring the technical condition of diesel engines in industrial or controlled laboratory conditions.

The technical roadmap of this scoping review is presented in Figure 1.

2. Materials and Methods

The methodological approach for investigating vibration methods for monitoring the technical condition of diesel engines was developed based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) principles, considering their adaptation to the objectives and specifics of a scoping review [17,18].

The research procedure was structured in accordance with the requirements of reproducibility, transparency, and methodological consistency at all stages [19]. It included clearly formulated publication selection criteria, systematic searching across several databases, and stepwise filtering of materials. The search was conducted in three consecutive stages:

• identification of records via automated queries to databases;
• preliminary screening of titles and abstracts according to the inclusion and exclusion criteria;
• assessment of full-text materials considering their thematic relevance.

Studies describing technical implementations or experimental work where vibration signals were used for diagnosing faults or monitoring operational conditions of diesel engines were eligible for inclusion. Publications focusing on other monitoring methods, dedicated to non-diesel engines, or lacking sufficient methodological transparency were excluded. Only peer-reviewed English-language articles published between 2015 and 2025 were considered, allowing a focus on developments of practical significance from the last decade.

Primary screening was conducted by two reviewers independently of each other. Disagreements or doubts regarding inclusion were resolved through discussion, with the involvement of a third expert if necessary to reach a consensus.

In accordance with the objectives of a scoping review, this study did not perform a formal critical appraisal or quantitative ranking of the quality of the included publications. Instead, all selected works underwent thematic categorization and analysis to identify research trends, methodological diversity, and areas requiring further development to advance the practical implementation of vibration diagnostic systems into the real-world operating conditions of diesel engines.

To ensure transparent and complete reporting, we assessed this review against the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR). The completed PRISMA-ScR Checklist is available in the Supplementary Materials (Table S1) and, for each item, specifies the exact manuscript page(s) where it is addressed. Items not applicable to this scoping review are marked “not applicable.”

2.1. Protocol and Registration

Prior to commencing work on the review, a detailed methodological protocol was prepared, formalizing the study’s structure and procedures. The protocol was developed by the research team in accordance with the PRISMA-ScR principles [17], which provide methodological guidance specifically for scoping reviews. The document clearly defined the objectives, research question, the Population–Concept–Context (PCC) structure, inclusion and exclusion criteria, databases for the search, article selection strategy, and data extraction procedures.

Given that international registries (e.g., OSF) currently do not accept scoping reviews for official registration, the protocol was not publicly posted. Nevertheless, the final version of the document underwent internal review and was approved by the research supervisors before data collection began. All steps of the procedure, including adjustments to inclusion parameters and the logic for data synthesis, were documented to ensure transparency. A complete description of the methodological process is provided in Section 2.2, Section 2.3, Section 2.4, Section 2.5, Section 2.6 and Section 2.7 of the article.

2.2. Eligibility Criteria

The selection criteria were defined in advance to ensure consistency and transparency at all stages of the selection process. The goal was to include studies capable of making a substantive contribution to the structured analysis of vibration monitoring methods for the technical condition of diesel engines. The inclusion and exclusion criteria were aligned with the scope and objectives of the review.

• Inclusion Criteria.

Publication Period: Studies published from 1 January 2015 to 18 July 2025.

Language: English-language publications only.

Document Type: Peer-reviewed journal articles.

Thematic Focus: Studies using vibration data for the detection, diagnosis, or monitoring of diesel engine faults.

Application Context: Studies conducted in laboratory or industrial settings.

• Exclusion Criteria.

Papers focused on non-diesel engines (e.g., gasoline or electric).

Articles where vibration analysis was not applied or was used exclusively for solving irrelevant tasks (e.g., fatigue strength modeling, improving acoustic comfort) without application to condition monitoring or diagnosis.

Studies not available for full-text review at the time of the full-text screening.

2.3. Information Sources

To ensure comprehensive coverage of the literature on vibration monitoring of the technical condition of diesel engines, a structured search was performed across five bibliometric databases: DOAJ (Directory of Open Access Journals), PubMed, ScienceDirect, Scopus, and Web of Science. All identified records were exported to EndNote 25 for the removal of duplicates, labeling, and subsequent screening.

2.4. Search Strategy

A reproducible, structured search strategy was applied to identify peer-reviewed studies describing the application of vibration methods for condition monitoring, failure detection, or diagnosis of diesel engines. The strategy was iteratively adapted to the specific syntax of each platform (accounting for differences in indexing, functions, and filters). The search was performed in the Title and Abstract fields across all five databases.

The core Boolean search string was: (“vibration monitoring” OR “vibration analysis” OR “vibration condition monitoring” OR “vibration fault detection”) AND (“diesel engine” OR “compression ignition engine”).

This formulation was intentionally designed for inclusivity and to maximize search recall: it encompasses literature on vibration methods in the context of diesel engine diagnostics without restricting the search to specific analysis tools (e.g., Fast Fourier Transform, machine learning methods, wavelet approaches), which vary significantly across studies.

The applied search filters were:

• publication type—journal article;
• years of publication—2015–2025;
• language—English.

2.5. Study Selection

The selection procedure followed a structured multi-stage approach in accordance with the PRISMA-ScR recommendations and was designed to ensure methodological transparency and reproducibility. The complete workflow is shown in Figure 2 using the PRISMA 2020 flow diagram for systematic reviews of databases and registries.

• Identification.

A total of 510 records were identified from searches across five bibliometric databases: DOAJ, PubMed, ScienceDirect, Scopus, and Web of Science, for the period from 1 January 2015 to 18 July 2025. Duplicates (n = 122) were automatically detected and removed using EndNote 25, leaving 350 unique records for screening.

• Screening.

All 350 records were screened based on their titles and abstracts. At this stage, publications that were incidentally included in the search results were excluded. Specifically, works where the study focused on a non-diesel engine were removed. For example, studies examining, for instance, an agricultural pneumatic loosener and its vibrations in general, rather than the vibrations of a diesel engine, were identified and excluded, as this scoping review fundamentally requires the analysis of diesel engines and the vibrations arising within them. Publications where vibration analysis was only mentioned as one possible condition monitoring approach but was not actually investigated were also excluded. At this stage, 97 records were excluded. Consequently, 253 records were retained for full-text retrieval and subsequent evaluation.

• Full-Text Eligibility Assessment.

Out of the 253 publications, full texts were obtained and assessed for 185 publications. Based on the full-text eligibility assessment, 108 publications were excluded. The reasons for exclusion were:

• vibration used as an auxiliary parameter in fuel-related studies (n = 43);
• engineering and design studies without monitoring objectives (n = 29);
• algorithmic, methodological, or theoretical works without applied validation (n = 17);
• post-failure diagnostics (n = 13);
• other irrelevant cases (focus on acoustics/noise or on components not related to diesel engine vibration monitoring tasks) (n = 6).

The entire process of selecting relevant sources is illustrated in Figure 2 (PRISMA 2020 flow diagram), which details the number of records at each stage and the reasons for exclusion, thereby ensuring transparency of the decisions made.

2.6. Data Charting, Management, and Items

To ensure the completeness and reliability of the evidence synthesis, data mapping was performed systematically using a standardized and calibrated extraction form. A structured mapping form was prepared in Microsoft Excel and used for the uniform extraction of key study characteristics and their results.

Table 1. Standardized data mapping form and list of extracted variables.

Category	Breakdown	PCC Relation
Publication Details	Authors, Year, Journal	-
Study Characteristics	Geography, Sector of Application (Power Generation, Transportation, Marine, Mining, Other)	Population/Context
Research Focus	Monitoring Methods, Sensor Types and Locations, Diagnosed Components/Subsystems, Digital Technologies Used, Stated Limitations	Concept
Key Results	Main Findings; Reported Advantages and Limitations of the Approach	Concept/Context
Methodological Details	Applied Signal Processing Methods; Applied Analytical/Diagnostic Models; Software Tools Used	Population/Context

presents the complete mapping form. It lists the variables that were systematically extracted from each included study in accordance with the review’s objectives. These items reflect the Population-Concept-Context (PCC) framework.

2.7. Data Synthesis

Data synthesis of the included studies was performed through structured thematic analysis aimed at identifying recurring patterns, contextual differences, and technological trends. Relevant aspects are summarized in

Table 2. The three aspects and related sub-categories of the reviewed studies.

Application Aspect	Diagnostic and Functional Targets	Digital and Modeling Technologies
General industrial installations; Transport; Power generation; Marine and ship propulsion systems	Air management systems; bearings; combustion dynamics; crankshaft torsional vibrations; rotational speed fluctuations; cylinders; fuel injection; general technical condition; lubrication system; pistons; rotating assemblies; supporting structures; turbocharging; valve train; vibration transmission paths	Classical signal processing; machine learning; statistical analysis and threshold-based alarm algorithms; deep learning; conceptual monitoring frameworks; numerical modeling; large language models; hybrid/integrated architectures

, and the results of the analysis of the studies are presented in Appendix A (Table A1).

The synthesized findings were aligned with the research objectives to assess the level of technological advancement, identify limitations (such as sensor cost, data noise, etc.), and track trends towards automation and real-time monitoring. This integrated consideration made it possible to outline the key contributions of existing studies and highlight areas requiring further investigation.

3. Results

3.1. Research Scope

Over the past decade, a significant body of work on vibration monitoring and diagnostics of diesel engines has emerged. The literature review shows a strong growth in publications in the early 2020s compared to previous years: 29 publications for the period 2015–2019, compared to 48 publications from 2020 to 2025. The number of publications is increasing, and numerous studies with increasingly advanced methods and expanded application areas appear annually. Geographically, the research is global, yet a concentration of work in Asia and Europe is noticeable. Chinese researchers are particularly productive, systematically publishing studies on engine vibration diagnostics, including D. Zhen [20], S. Ji [21], C. Zhao [22], H. Bai [23], A. Wang [24] et al. European groups also make substantial contributions, notably from Italy [25,26,27] and Poland [28,29,30]. Results from other regions are also noted: Iran [31], Malaysia [32], Japan [33,34,35], and the Republic of Korea [36,37]. This diversity reflects the universal nature of diesel monitoring tasks and the high applicability of vibration methods.

In terms of application sectors, transport predominates—mainly automotive and locomotive diesel engines. A substantial body of research focuses on marine propulsion systems, where the highest-power engines are low-speed two-stroke machines; their operating specifics necessitate specific monitoring approaches [38,39]. Another notable share is focused on stationary power generation [40] and general industrial diesel engines (including those used in agricultural machinery and military applications) [41], confirming the applicability of vibration diagnostics to generator sets and heavy machinery. The scale of the studied objects varies from small diesel engines for micro-cars [42] and light automotive units [26] to large multi-cylinder marine engines [43], demonstrating the validation of methods across different power levels.

Methodologically, the vast majority of studies use the measurement of structural vibrations as the primary diagnostic channel. In this review, structural vibrations refer to accelerations measured on the engine structure using surface-mounted accelerometers. Non-invasive vibration measurements on engine components—the cylinder block, head, crankshaft—have become standard practice for capturing the system’s dynamic response. The predominance of vibration monitoring is explained by its sensitivity to a wide range of faults and combustion process deviations, as well as the relative simplicity of sensor installation. A number of studies propose enhancing information content through the use of multimodal schemes. Here, vibration is combined with acoustic/noise measurements [20,44], and in some cases, also with a direct in-cylinder pressure sensor for validating combustion phases [45]. Such hybrid approaches aim to correlate surface vibration signatures with indicator pressure fluctuations or acoustics and improve the reliability of defect detection.

Research activity has accelerated since 2020 and is globally distributed—with the strongest concentration in Asia and Europe—and applications are dominated by transport, followed by marine propulsion and stationary power generation across a wide range of engine sizes. Methodologically, surface-mounted accelerometers measuring structural vibrations remain the primary non-invasive method, while multimodal setups (vibration with acoustics and, in some cases, in-cylinder pressure) are increasingly used to validate combustion phasing and improve diagnostic reliability.

3.2. Measurement Tools and Sensor Configurations

In nearly all analyzed studies, the primary diagnostic method is the recording of structural vibrations using accelerometers mounted directly on the engine casing. Piezoelectric accelerometers are typically used for this purpose, mounted on the cylinder head or engine block to record mechanical responses caused by combustion and the movement of moving parts. In several studies, sensors were installed on the cylinder head to measure vibrations directly caused by combustion [21,46,47]; in other cases, on the engine block [40] or intake manifold [31], depending on the diagnostic objectives. The accelerometers used generally have a wide frequency bandwidth, enabling the recording of sharp pressure impulses and resonant phenomena accompanying the combustion process. For instance, one study used six highly sensitive piezoelectric accelerometers (BW14100) with a frequency range of 1–8 kHz, evenly distributed over the engine surface, for comprehensive vibration analysis [48]. In another case, a triaxial accelerometer PCB-ICP356A16 with a frequency range of 0.5–5 kHz, mounted on the engine block, was used [49]. The practice of placing multiple accelerometers at various points and in different directions is common, providing a more complete picture of vibrational processes in the engine [25,50].

Alongside vibration sensors, many studies employ piezoelectric pressure sensors installed directly in the cylinders. Such sensors are typically integrated into modified glow plugs or installed via threaded connections, for example, the water-cooled pressure sensor Kistler 6052C [33] or the quartz transducer AVL GU13P [45]. In-cylinder pressure provides reference values for key parameters: start of combustion, peak pressure, and pressure rise rate, against which results obtained from vibration signals are verified [30]. Synchronizing pressure and vibration time series allows for reliably establishing the correspondence between extracted vibration signal features and the actual combustion process.

A separate group of studies supplements structural vibration measurements with airborne acoustic signals recorded using microphones placed near the engine. Specifically, high-quality condenser microphones are used for this, such as the “Bruel & Kjaer” Type 4939 [45], capable of recording engine acoustic emissions. Microphone placement is crucial, as engine acoustic fields are highly directional and susceptible to external noise. For instance, in the study by Narayan S. [44], a microphone placed near the cylinder block primarily captured combustion noise, while another mounted at the intake manifold recorded acoustic oscillations related to intake. This example demonstrates that multi-point acoustic measurements allow for selectively analyzing different engine noise sources.

For diagnosing engine rotational dynamics, several studies employ methods for measuring torsional vibrations and the instantaneous angular speed of the crankshaft. Such measurements are implemented using high-precision optical encoders or laser tachometers capable of recording minute speed deviations in each cycle. The obtained data reflect torsional vibrations arising from alternating power strokes and can indicate power imbalance between cylinders, misfires, or mechanical issues in the drivetrain. For example, in the work by Drewing et al. [29], a laser encoder with a sampling frequency of 16 MHz was used to record the instantaneous crankshaft rotational speed; analysis of the torsional spectrum allowed for detecting injector faults based on harmonic anomalies. Other studies used strain gauges attached to the shaft or engine flexible coupling, enabling direct measurement of torque and its fluctuations [37,51]. For instance, in the work by Palomo Guerrero [38], the installation of strain gauge type sensors allowed, using a trained model, the authors to estimate the indicated power of cylinders with an error of less than 1%. Despite their high diagnostic value, implementing such measurements involves technical difficulties—low signal amplitude, the need for high installation and calibration accuracy, and the application of specialized signal processing methods [52].

In recent years, a trend towards developing portable and wireless diagnostic devices intended for integration into the engine system without using bulky laboratory equipment has been observed. Such solutions include compact modules with accelerometers and additional sensors (e.g., for temperature or pressure), equipped with telemetry and onboard data processing capabilities. Specifically, Kucherenko et al. [43] developed a Bluetooth module combining a piezoelectric in-cylinder pressure sensor and a vibration sensor. Such a device allows for real-time tracking of injection timings and valve opening events with information output to a mobile application. Another study proposed using an inexpensive multi-sensor recorder including accelerometers, microphones, thermocouples, and gas analyzers for comprehensive engine condition monitoring [53]. Despite their promise, such solutions mostly remain at the prototype stage and require further validation under long-term operation and conditions involving vibration, temperature fluctuations, and electromagnetic interference [54]. In our view, these devices remain pre-commercial because real-world deployment simultaneously demands long-term mechanical/thermal endurance, strong EMC immunity in electrically noisy engine bays, and tight power budgets despite the bandwidth/latency and interference limits of wireless links for multi-channel kHz vibration data—constraints that are hard to satisfy at small size and cost. On top of that, maintaining cross-sensor time-synchronization and calibration stability over months, plus meeting hazardous-area certifications and modern IIoT cybersecurity/maintenance requirements, lengthens qualification cycles; most designs stay at prototype/pilot until these are solved in one integrated package.

Analysis of research results shows that a substantial number of studies use multi-point and triaxial configurations to improve diagnostic reliability. Combining vibration signals with cylinder pressure data and acoustic measurements is used to refine combustion phases and identify knocking. Measurements of torsional vibrations and instantaneous angular speed are used less frequently but provide higher sensitivity to imbalance and misfires.

In sum, the practical baseline remains surface-mounted piezoelectric accelerometers—often in multi-point/triaxial layouts with bandwidths in the kHz range—to capture combustion-driven impulses and resonances. Complementary channels (in-cylinder pressure as a phasing reference and strategically placed microphones for airborne acoustics) improve reliability, while torsional/instantaneous-speed sensing offers higher sensitivity to imbalance and misfires at the cost of greater installation and processing complexity. Portable/wireless multi-sensor modules are promising but largely pre-commercial due to durability, EMC, power/bandwidth, synchronization, certification, and cybersecurity constraints; thus, the most deployment-ready solution currently is wired piezo arrays with planned placement, synchronization, and optional pressure/acoustic references.

3.3. Diagnostic and Functional Objectives

The most developed area of vibration monitoring for the technical condition of diesel engines is the non-invasive extraction of combustion parameters from vibration signals. Casing oscillations are used to estimate the start of combustion, peak pressure position, and maximum pressure rise rate [20,40,46]. The verified accuracy for phase determination is around 0.5–2 degrees of crank angle (°CA) relative to indicator pressure [21,47,55], with specific vibroacoustic indices consistently correlating with cycle-by-cycle thermodynamic metrics under different injection strategies [45]. Vibration reliably detects knocking conditions in dual-fuel diesel engines (the knock index aligns with indicator diagram data) [30,32].

Diagnostics of the injection system relies on characteristic changes in the vibrational signature across cycles. Classification of misfires and misfiring events achieves >97.5% accuracy in laboratory conditions [31,48,56], and the use of the Modified S-Transform (MST) and Two-Dimensional Non-negative Matrix Factorization (2D-NMF) for feature extraction enables the classification of injection faults based on vibration signals [57]. As research [28] showed, diagnostic features of injector cracks can be detected 247 h before their failure. This lead time represents an extreme case under controlled laboratory conditions; most detections occur within the last five days before failure. Injector coking is identified by anomalies in the torsional vibration spectrum [29]. For multi-cylinder engines, accuracy close to 100% in localizing the faulty cylinder in two-step schemes has been demonstrated [58], and failures of the fuel rail pressure (FRP) sensor and mass air flow (MAF) sensor are detected by changes in vibration harmonics [59].

The condition of the valve train is diagnosed based on the impact components of cylinder head vibration. The use of Wavelet Packet Transform (WPT) and Mel-Frequency Cepstrum Coefficients (MFCC) ensures robust detection of excessive valve clearance [60,61,62]. Sparse compression and dictionary models enable wireless data transmission without loss of diagnostic information [63,64].

Defects in the cylinder-piston group manifest in mid- and high-frequency ranges: for incipient scuffing, a spectral “band” of 2.4–4.7 kHz with a maximum at 3.7 kHz is characteristic. The energy of piston slaps concentrates around 5 kHz and decreases with increasing speed, whereas chamber resonances at 6–8 kHz are weakly dependent on frequency [34]. Variational Mode Decomposition (VMD) combined with deconvolution allows reconstructing the impact force profile versus crank angle [65], and Dynamic Principal Component Analysis (DPCA) combined with VMD detects early piston pin wear not detectable by standard spectral wavelet analysis [66]. Experiments by Ramteke et al. [67] confirm the correlation between increased vibration/noise and decreased power, worsened brake thermal efficiency (BTE), exhaust emissions, and oil degradation. The influence of lubrication on the vibroactivity of cylinder liners is also quantitatively demonstrated [68]. During ring scuffing, the second harmonic of torsional vibrations increases sharply (>5 times)—an early sign of failure [69].

Detection of bearing and gearbox defects is achieved through informative time/entropy features and improved signal-to-noise ratio (SNR) via adaptive filtering: high accuracies up to ≈99.16% and noise robustness have been demonstrated [70,71,72]. Using Ensemble Empirical Mode Decomposition (EEMD) for feature extraction from vibration signals followed by classification using a Directed Acyclic Graph Support Vector Machine (DAGSVM) successfully recognizes typical gearbox faults [24].

Crankshaft dynamics are assessed via Instantaneous Angular Speed (IAS) and torsional vibrations. This allows for quantitative diagnosis of misalignment [73], imbalance [39], and degradation of flexible couplings (online identification of stiffness, error < 5%) [51].

Integrated and portable solutions combine vibration with acoustics/temperature/gas analysis and Internet of Things (IoT) telemetry. The literature describes: an energy-autonomous accelerometer based on a Freestanding Triboelectric Nanogenerator (F-TENG) with 97.87% accuracy [22], an inexpensive multimodal logger for real-time monitoring [53], optimal microphone configurations for separate recording of combustion and intake noises [44], a portable module for monitoring pressure and vibration with onboard computation of Top Dead Center (TDC) and Cycle Irregularity Index (CII) [43]. Additionally, accelerometers on the compressor housing allow measuring turbocharger speed and tracking its relationship with combustion changes [25,26].

Thus, vibration data reliably assess combustion phases and quality (including knock and emission markers), diagnose injection faults, valve and piston defects, and rotational dynamics disturbances with high accuracy. Tasks related to combustion and general engine condition assessment are the most significant. A clear trend towards using integrated multimodal IoT systems and portable models suitable for in-service monitoring is evident.

Overall, vibration signals provide a non-invasive route to combustion phasing and quality with degree-level crank-angle precision, while knock and injection faults—including early injector-crack precursors—are reliably captured; torsional/instantaneous-speed analysis adds sensitivity to imbalance, misalignment and coupling degradation. Valve-train impacts and piston/liner defects are resolved by targeted time–frequency methods (e.g., WPT/MFCC, VMD/DPCA) and characteristic mid/high-frequency bands, and bearing/gearbox issues are separated using informative features with high reported accuracies. A clear trend is toward integrated multimodal setups and portable IoT modules (vibration with pressure/acoustics and auxiliary sensing), extending these diagnostics from test benches toward in-service monitoring.

3.4. Digital Technologies and Models Used in Vibration Monitoring

As the literature review shows, classical time-frequency methods are widely used in research: Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), Continuous and Discrete Wavelet Transforms (CWT; DWT), Wigner-Ville Distribution (WVD), as well as adaptive decompositions. These methods allow for isolating vibration components related to combustion and impact processes and analyzing them across frequency bands. Wavelet analysis and cross-wavelet correlation enable the identification of delays and vibration transmission paths from the cylinder [25]. The Wigner-Ville Distribution and Fractional Fourier Transform (FRFT) are most preferred for investigating combustion noise structure and assessing fuel quality [36]. Empirical and Variational Mode Decompositions (Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD)) provide adaptive extraction of Intrinsic Mode Functions (IMFs) and precise identification of cycle parameters [55].

For condition monitoring, statistical characteristics and threshold-based deviation criteria are also used; the standard deviation, variance, and derivatives of the vibration signal are tracked, and empirical indices characterizing the combustion mode and type are introduced. Q-control charts (Q-charts) reliably record anomalies, as confirmed during ship operation [50,74]. For knock detection, an integral knock index with a three-sigma rule threshold, consistent with pressure sensor data, is used [32].

Classical machine learning algorithms are applied for automatic classification and regression using vibration data; the most widespread are Support Vector Machines (SVM), k-Nearest Neighbors (KNN), and simplified neural networks: Extreme Learning Machines (ELM), Multilayer Perceptrons (MLP), etc., with mandatory informative feature extraction. The combination of multiscale permutation entropy with KNN enables the determination of engine and bearing states [70]. SVM models allow identifying key combustion parameters (Start of Combustion (SOC), Maximum Pressure Rise Rate (MPRR), etc.) from vibration signals with up to 99.72% accuracy [31]. The application of ELM combined with adaptive decomposition improves vibration prediction accuracy and opens the possibility for effective real-time condition assessment [75].

Deep learning methods automatically extract hidden patterns from vibration time series and spectrograms. Convolutional networks and their hybrids are the most effective [76]. One-dimensional Convolutional Neural Networks (CNN) combined with Long Short-Term Memory (LSTM) reliably classify operating regimes, and adaptive pruning improves generalizability to another engine [77]. Architectures with a feature residual compression mechanism demonstrate 98.38% accuracy and noise robustness, outperforming classical approaches [78]. The Vision Transformer proved effective in analyzing synchronized spectrograms: using transfer learning and domain adaptation ensures high recognition accuracy (98.31% on published data, 95.67% on laboratory data) when transferring between engines and regimes, reducing the need for large datasets [79].

For cross-engine deployment, domain adaptation aims to reduce the source–target feature-distribution discrepancy so that a model trained on one engine generalizes to another. In practice, the training objective is augmented with a discrepancy penalty that drives multi-domain feature alignment while the standard label-prediction loss preserves task-relevant separability. In vibration diagnostics, DA couples with impact-component decomposition of vibration signals to obtain more transferable representations under changes in load regime and sensor placement, thereby reducing target-domain annotation needs and stabilizing decision boundaries when the operating envelopes are comparable [80].

Hybrid schemes combine preliminary signal transformation and learning; here, DWT/WPT, VMD, and other filters are used for feature extraction followed by classification. Multivariate VMD with band energy analysis combined with SVM enables defect recognition with >99% accuracy [81]. For wireless systems, compression and recovery methods are in demand: compressive sensing and sparse coding reduce data volume while preserving diagnostic information content [63]. Stacked Sparse Autoencoders (SSAE) enable the extraction of compact features; in combination with SVM, an accuracy exceeding 98% is achieved with fewer sensors and features [48].

Improved accuracy is provided by model optimization and combination: Genetic Algorithms, Particle Swarm/Firefly Algorithms, etc., are used for hyperparameter tuning and feature selection. Genetic Algorithms improve feature selection after adaptive decomposition and the quality of state clustering [56]. Tuning SVM with a Firefly Algorithm combined with improved fuzzy entropy increases recognition accuracy to 98.2% [41]. Multi-kernel Relevance Vector Machines (RVM), optimized by swarm methods, show advantages over traditional SVM and enable real-time operation [82]. Ensemble CNNs with aggregation based on Dempster-Shafer theory enhance robustness and accuracy compared to single networks [83].

Physics-based models complement digital processing, improving result interpretability. For example, an Autoregressive Moving Average (ARMA) model with Fourier harmonics describes the cyclic nature of vibrations in a 4-cylinder, 4-stroke diesel engine and generates angle-frequency spectra comparable to the Smoothed Pseudo Wigner-Ville Distribution (SPWVD) [84].

To facilitate a side-by-side comparison of applicability, advantages, and limitations,

Table 3. Comparative overview of methods for vibration-based engine condition monitoring.

Method Family	Typical Inputs/Representation	Data Needs	Latency/Compute & Real-Time	Robustness to Noise/Non-Stationarity	Interpretability	Typical Use Cases
FFT/Order analysis	Time series → spectra/orders (RPM-ref when available)	Low–moderate	Very low; on-board/edge friendly	Best on steady/tonal content; sensitive to drift	High (clear lines)	Baseline spectral health, order content, imbalance hints
STFT	Sliding spectra (time–freq grid)	Moderate	Low–moderate; edge feasible	Handles mild transients via windowing	High–moderate	Cycle-synchronous content, combustion pulse tracking
Wavelets	Scalograms/packet bands	Moderate	Moderate; edge feasible with selected bands	Good on impacts/bursts; tolerant to non-stationarity	Moderate	Combustion cues, valve-train impacts, knock bands
WVD/FRFT/SPWVD	High-res time–frequency/chirp-aligned views	Moderate	Moderate–high; often offline/edge-accelerated	High resolution; cross-term artifacts possible	Moderate	Combustion-noise structure; fuel quality/combustion regime studies
Adaptive decompositions	IMFs/modes; mode energies & envelopes	Moderate	Moderate–high; edge feasible for tuned VMD	Good for mixed, non-stationary signals; prone to mode mixing	Moderate	Isolating impulsive modes; deconvolution; early defect cues
Statistical & thresholds (incl. Q-charts)	Cycle/window features; control limits	Low	Very low; on-board ready	Robust if features stable; sensitive to setup drift	High	Simple anomaly flags; shipboard/field trend monitoring
Classical ML	Hand-crafted features (time, frequency, time–freq, entropy)	Moderate	Low–moderate; edge on CPUs/MCUs	Stable if features robust across regimes	Moderate (feature-level)	Injection/misfire classification; SOC/MPRR estimation
Deep learning	Raw time series or spectrograms	High	High (train & often infer); edge needs accelerators/pruning	Strong on matched data; needs TL/DA for transfer	Lower (requires XAI)	Multi-fault classification; regime recognition; high lab accuracy
Hybrid pipelines	Selected bands/modes → classifier	Moderate	Moderate; edge feasible	Good noise rejection via pre-selection	Moderate–high	Defect recognition with compact models; balanced accuracy/latency
Compression & sparse coding	Compressed features/sparse codes	Moderate	Shifts compute on-sensor; reduces link load	Preserves salience if task-aligned	Moderate	Wireless CM with reduced data volume
Physics-informed	Angle-sync/parametric fits	Low–moderate	Low–moderate; on-board feasible	Robust when assumptions hold	High	Angle-frequency structure, torsional dynamics, interpretability
Transfer/Domain adaptation	Source + target features/embeddings	Target labels: none–few	Higher train cost; inference unchanged	Improves cross-engine/regime generalization	Depends on base	Cross-engine transfer; less target annotation

summarizes the method families (inputs, data needs, latency/compute, robustness, interpretability, and typical use cases).

3.5. Limitations of Using Vibration Monitoring for Diesel Engines

One of the key limitations of vibration monitoring for diesel engines is the difficulty of separating signals from different vibration sources. In a multi-cylinder engine, the vibroacoustic response is a combination of oscillations from multiple cylinders and components that overlap. For instance, oscillations caused by fuel combustion, piston assembly movement, and the operation of valves and injectors manifest in overlapping frequency ranges, leading to the overlapping of spectral components [34,80]. This complicates the localization of a specific fault source, for example, identifying the problematic cylinder from the vibration signal [58]. Even the application of advanced signal separation methods does not guarantee complete separation of superimposed vibration signals, especially in the presence of external noise [78,85].

Another common problem is the high degree of noise and non-stationarity of diesel engine vibration signals. The oscillatory processes in an engine are distinctly nonlinear and non-stationary: the spectral composition of vibration can change significantly from cycle to cycle, complicating the extraction of stable diagnostic features [71,75]. Under strong interference, traditional frequency methods (FFT, spectral density estimation) may be insufficiently informative: different types of faults can produce similar spectral characteristics, masked by the noise background [21,86]. Thus, extracting reliable fault features against a background of non-stationary, multicomponent signals remains a non-trivial diagnostic task.

Many modern vibration diagnostic techniques impose high demands on the setup and computational resources of the hardware used. Methods for empirical signal decomposition and their improved variants require careful selection of decomposition parameters (e.g., the number of extracted components); otherwise, situations like mode mixing or excessive splitting of the signal into components occur [75]. High time and frequency resolution (as achieved, for example, using the Continuous Wavelet Transform) comes at the cost of a sharp increase in computational complexity; hence, in practice, one often has to resort to faster but coarser analysis methods (such as the Short-Time Fourier Transform) [87]. An excessive set of features is also undesirable: increasing feature dimensionality leads to longer processing times and the risk of overfitting diagnostic models [70]. Furthermore, complex machine learning algorithms (e.g., neural networks) are critically dependent on the volume and quality of data: with an insufficient training dataset, the model can get stuck in uninformative local minima, failing to correctly identify characteristic fault features [61].

Significant difficulties are also associated with the technical implementation of monitoring sensor systems. Reliable diagnostics often require the installation of multiple accelerometers and other sensors on the engine; however, selecting informative measurement points and maintaining such a sensor complex under real-world conditions is challenging [28,47]. Some important parameters are difficult to measure directly: for instance, monitoring indicated pressure or crankshaft torsional vibrations requires expensive built-in sensors, whose installation is not always feasible on an operating engine [29]. The implementation of wireless systems is limited by the bandwidth of data transmission channels: high-frequency vibration signals are problematic to transmit without loss, necessitating data compression and simplification of analysis algorithms [63]. Additionally, multi-channel measurement systems require careful calibration and synchronization; even minor sensor misalignments or external vibrational interference can distort the recorded signal, complicating its interpretation [88].

Another notable limitation is the insufficient testing of vibration analysis methods under real-world conditions and on different types of diesel engines. A significant portion of research has been conducted on engines in laboratory settings, and algorithms are typically validated under steady-state operating conditions (constant load and speed), whereas in reality, engines operate under variable loads and transient conditions. The influence of external factors (fuel quality, marine operating conditions, vibrations from movement, etc.) is often not accounted for, although in practice such influences can significantly affect signal parameters [50]. All this points to the necessity for broader validation of proposed approaches—long-term field trials under various conditions on different types of diesel engines and under variable regimes are required to confirm reliability and transferability.

4. Discussion

The analysis of 77 publications showed that vibration monitoring methods are currently widely used for diagnosing diesel engines; however, their implementation in industrial settings remains limited. Most studies have been conducted under laboratory conditions, whereas work under real-world operating regimes of diesel engines is scarce and accompanied by additional difficulties—data transmission, noise interference, and signal instability. Thus, the reliability of approaches to vibration monitoring of the technical condition of diesel engines during operation requires further verification.

The most common solution here is the recording of engine structural vibrations using piezoelectric accelerometers installed on the cylinder block or head. This method allows for determining combustion phases with an accuracy of 0.5–2° of crank angle and detecting injection system faults with high reliability. Combining vibration data with pressure or acoustic data enhances diagnostic reliability.

Beyond piezoelectric accelerometers, two sensor families merit attention for harsh environments: fiber Bragg grating sensors and micro-electromechanical systems (MEMS) devices. FBG sensors are inherently immune to EMI, tolerate high temperature gradients, and support long-run multiplexing—traits attractive for engine compartments—though interrogator cost and ruggedized packaging under vibration remain practical barriers to wide adoption [89,90]. Modern MEMS accelerometers and ultrasonic MEMS microphones enable compact, low-power nodes suitable for embedded or wireless configurations and have shown accuracy comparable to piezo solutions in rotating-machinery tests; nonetheless, usable bandwidth/noise floor in the kHz range and temperature-induced drift call for careful calibration and periodic re-verification [91,92].

Regarding the algorithms used for vibration diagnostics, the foundation consists of time-frequency transformations (FFT, STFT, CWT), supplemented by adaptive methods (EMD, VMD) and machine learning (SVM, KNN, MLP). Modern deep learning models (CNN, LSTM, transformers) in some studies achieved accuracy around 98% [22,78]; however, these results were obtained under controlled laboratory conditions and are not confirmed for industrial engines functioning during their actual operation.

Limitations are associated with the multi-component nature of vibration signals, overlapping spectra from different mechanisms, non-stationarity, and high noise levels. Reliable diagnostics require the placement of multiple sensors, which complicates the system, and access to parameters like torsional vibrations remains technically challenging. Wireless solutions face limitations in channel bandwidth and the necessity for preprocessing vibration signals at the sensor installation site.

Thus, the research results indicate the high efficiency of vibration diagnostics for diesel engines under laboratory conditions, but its transfer to real-world engine operating conditions requires overcoming significant barriers. Promising directions here are:

• Conducting field trials and model validation on industrial installations;
• Creating multimodal sensor platforms (vibration, pressure, acoustics) with wireless data transmission;
• Developing noise-resistant algorithms and domain adaptation methods;
• Optimizing computational resources for real-time analysis;
• Accounting for external factors (fuel quality, transient regimes, climatic conditions).

In general, further progress in vibration monitoring of the technical condition of diesel engines requires an interdisciplinary approach combining new sensor solutions, processing algorithms, and IoT technologies, which will enable the creation of reliable systems.

A typical deployment scenario for a diesel generator set, intended for supplying electricity to consumers, can be outlined as follows. The minimal configuration includes 2–4 vibration sensors on the head and/or cylinder block near the injectors and support bearings, one acoustic sensor in the engine compartment (shielded from airflow), and a rotational speed reference signal (a tachometer signal from the generator or an angular sensor). Signal preprocessing from the vibration sensors is performed locally: synchronization based on RPM, bandpass filtering within windows of expected orders, detection of combustion pulse envelopes, suppression of outliers; subsequently, simple cycle-based features are calculated (energy, amplitudes of dominant orders, relative position of pulses by crankshaft angle) and threshold maps or a compact machine learning model are generated. Only the features and events are transmitted through the communication channel, while the raw data is stored locally for subsequent analysis. Approximate performance indicators are: a false positive rate of no more than a few percent per day; detection of misfires and injection faults within one or two operating cycles; estimation of combustion phases with an error of roughly several degrees of crank angle (given angular referencing). The requirements for the communication channel when transmitting events are at the level of tens of kilobits per second; the computational load corresponds to the throughput of compact ‘field’ controllers. Commissioning includes creating a sensor placement map, securing mounts/adhesive bonds, routing cables considering electromagnetic interference, and recording reference operating regimes; the maintenance schedule involves periodic self-checks, monitoring for drift, and annual calibration. Specific values are adjusted according to the engine type, speed range, and layout. This approach already ensures reproducibility, transparent quality criteria, and meets the requirements for communication and computing.

5. Conclusions

The conducted review has systematized modern approaches to vibration monitoring of the technical condition of diesel engines and revealed a gap between laboratory research results and the requirements of industrial operation. The analysis included 77 peer-reviewed studies covering various application sectors and engine classes, which provided a representative picture of the methods used and their application scenarios.

The main technical conclusions are as follows. The predominant technical solution for obtaining information about vibration signal parameters is the recording of structural vibrations using piezoelectric accelerometers installed on the cylinder block or head; the combined use of vibration measurements with pressure data and acoustic parameters enhances diagnostic reliability. Several studies demonstrate the possibility of estimating combustion process parameters from the engine surface with an accuracy on the order of fractions of a degree of crankshaft angle. However, the highest accuracy values are typically obtained under controlled laboratory conditions. These observations are consistent with reviews of methods including time and frequency domain analysis, adaptive decompositions, and machine and deep learning algorithms.

Key limitations are both methodological and engineering in nature. These include: the multicomponent and non-stationary nature of the signals, a high level of noise, sensitivity of the results to sensor placement, difficulties in accessing torsional vibration parameters, and communication channel limitations requiring local preprocessing of vibration signals. A substantial portion of the work was performed on test benches and under steady-state conditions; confirmations of the robustness of the obtained results on real industrial installations and under their transient operating conditions are insufficient. These circumstances limit the transferability of the published solutions to industrial installations without additional validation.

The practical significance of the review lies in its identification of directions where the implementation of vibration monitoring systems is already feasible, but where additional efforts are required. In the short term, it is advisable to implement multimodal configurations (assessing vibration jointly with pressure and/or acoustic parameters) to improve the reliability of detecting injection faults and combustion process deviations; this should involve pre-designing the sensor placement scheme and calibration procedures, taking into account real-world noise and vibration transmission paths. In the medium term, it is necessary to conduct field trials on typical applications (power generation installations, transport, marine engines) involving prolonged monitoring and control of transient regimes.

The limitations of the review itself correspond to the scoping study format: a formal critical appraisal of the included publications’ quality was not performed; the considered period is limited to 2015–2025; the analysis covers only English-language articles. These boundaries reduced the risk of methodological heterogeneity but might have left some relevant developments outside the scope.

To bridge the lab–industry gap, future work should integrate engine digital twins [93,94] that fuse physics-based models with operational data, enabling closed-loop prognostics and health management—monitoring, diagnosis, prediction of remaining useful life, and maintenance/control actions with feedback—and safer transfer from test rigs to fleets. Deep models should be accompanied by explainable AI so that diagnostic decisions can be audited during operations [95,96]. Finally, standardized datasets, benchmarks and reporting are essential to compare methods fairly and to assess the cross-engine transferability [97].

Considering the obtained results, the priority directions for further research are:

• Multicenter field validations on industrial installations with strictly regulated protocols and independent benchmarks;
• Standardization of reporting and datasets (feature structures, metrics, validation scenarios) for result comparability;
• Development of noise-resistant algorithms and procedures for adaptation to specific operating environments;
• Design of measurement systems with explicit constraints on communication channel and computational resources, incorporating local preprocessing and managed data compression.

Implementing these steps will transform the efficiency of vibration monitoring methods for the technical condition of diesel engines, demonstrated in laboratory studies, into reliable and reproducible solutions for their industrial operation in modern power installations.