University of Southampton Projects

• Project type: Academic-led
• Supervisors: Sifeng Bi, Daniil Yurchenko

Vibroacoustic modeling is crucial in engineering systems for accurately predicting, monitoring, and controlling noise and vibration, especially in aerospace and automotive applications. Virtual decoupling techniques have been successfully applied to isolate and analyze vibroacoustic systems in low-frequency domain [1]. However, uncertainties often limit the model’s accuracy. Furthermore, vibroacoustic systems pose challenges due to the absence of a single analysis technique suitable for all frequency ranges [2]. Different physical behaviors and uncertainty impacts in low-, mid-, and high-frequency domains necessitate distinct approaches. Stochastic model updating methods combined with uncertainty quantification provide a systematic framework to address these issues and enhance the reliability of vibroacoustic models [3]. This project aims to integrate these methods, developing a comprehensive framework for the calibration, verification, and validation (V&V) of vibroacoustic models under uncertainty.

1. Bi, S., Ouisse, M., Foltête, E., & Jund, A. (2017). Virtual decoupling of vibroacoustical systems. Journal of Sound and Vibration, 401, 169–189.
2. Sestieri, A., & Carcaterra, A. (2013). Vibroacoustic: The challenges of a mission impossible? Mechanical Systems and Signal Processing, 34(1–2), 1–18.
3. Bi, S., Beer, M., Cogan, S., & Mottershead, J. (2023). Stochastic Model Updating with Uncertainty Quantification: An Overview and Tutorial. Mechanical Systems and Signal Processing, 204, 110784.

Objectives:

The objective is to develop a data-driven framework for vibroacoustic modeling across different frequency domains. This framework will integrate suitable combinations of techniques for low, mid, and high frequencies and incorporate dedicated model calibration and V&V approaches tailored to the specific features and uncertainty challenges of each frequency domain.

Proposed Work Packages:

WP1: Uncertainty Quantification (UQ) for Vibroacoustic Modelling

Tailor UQ techniques for each frequency domain: FEM and Modal Analysis for low frequencies, FEM-SEA and WFEM for mid frequencies, and SEA, Ray Tracing, and Geometrical Acoustics for high frequencies, addressing uncertainties specific to each domain.

WP2: Model Updating for Different Frequency Domains

Develop and apply model updating techniques for vibroacoustic systems across frequency domains, improving accuracy by incorporating experimental data and adjusting for uncertainties in each range.

WP3: Data-driven Approach for Efficiency and Applicability

Use data-driven techniques to enhance computational efficiency, reducing model complexity and making the framework more practical for real-world applications, especially in real-time settings.

WP4: Verification and Validation (V&V) Implement a rigorous V&V process by comparing model predictions with experimental data, ensuring accuracy across all frequency domains, and addressing specific uncertainty challenges for each range.

Academic-led project

Supervisors: Paul Kemp and Paul White

Project partner: Environment Agency

Globally, many fish populations have suffered dramatic declines over recent decades. This is most serious in fresh waters, which are the most degraded of all the worlds ecosystems. In rivers, the construction of dams and other structures negatively impact fish populations by blocking their movements, fragmenting habitat, and injuring those that enter water intakes, e.g. to hydropower turbines or water supply systems. Over many decades, a range of environmental impact mitigation technologies have been developed to help conserve river fish. For example, physical fish screens have been installed at water intakes to reduce entrainment and loss of fish. The effectiveness of such environmental impact mitigation solutions, however, can be highly variable between site, context, and species, and can result in unintended negative consequences as the physical screens themselves may cause injury and death due to mechanical abrasion and impingement. To improve the performance, behavioural devices may have a role to play when used in combination with traditional screening systems. Underwater sound has the potential to provide an effective deterrent because it is omnidirectional and so can reach multiple individuals simultaneously, while also effective under low illumination and turbid conditions. This project will advance acoustic deterrents for use combination with hybrid physical screens to help protect fish at water intakes and abstraction points. In particular, we will explore how acoustics may be used to repel multiple species in a community-based approach, rather than focusing on a few target species as is typically the case. The project will involve both experimental development and prototype testing in the field.

Academic-led project

Supervisors: Christine Evers and Jonathon Hare

State of the art deep audio models rely on models that are composed of billions of parameters and are trained on vast datasets. While training with huge datasets can provide generalizability in many settings, these models 1) lack interpretability, and 2) require prohibitive compute resources, and, hence, excessive carbon footprints.

The vision of this project is to shift focus from massive audio models and internet-scale datasets to embedding our understanding of the human auditory process in the model architecture. A wealth of auditory models that mimic the signal transformations along the auditory pathways exist in the signal processing and neuroscience communities. Most of these models rely on filters that are fixed and optimised a priori for specific listening tasks. While these models can be useful for pre-processing of data, the fixed nature of the filters may hamper generalisability, adaptivity, and overall performance of the resulting, trained models.

To enable generalisability across a variety of tasks, the aim of this project is to develop fully learnable auditory models. Specifically – with a view to enable sustainable AI – we want to incorporate in the model architecture differentiable filters that 1) require only few weights and 2) provide receptive fields at varying timescales. 

By developing models that reduce significantly the number of model parameters and, hence, required compute resources and training times, the project outcomes have significant potential for impact across a variety of applications that require the deployment of deep models onboard embedded devices, including robotics and immersive technologies. The project is directly aligned with two EPSRC-funded research programmes on which Dr Evers is either principal- or co-investigator: “Challenges in Immersive Audio Technologies (CIAT)” and “Active Audition for Robots (ActivATOR)”. As such, there is significant potential for collaboration and knowledge exchange with our project partners, including Stanford University (USA).

Academic-led project

Supervisors: Temistocle Grenga and Edward Richardson

Reduction of jet, fan and other turbomachinery noise sources in aeroengines leaves combustion noise as an increasingly important contribution to aircraft noise. Hydrogen-fuelled engines are being developed as a pathway to decarbonise aviation, however the impact of hydrogen combustion on noise emission is not well understood and there is a lack of reliable methods to predict associated sound emission.

This PhD research project aims to investigate the differences in sound generation between hydrogen and hydrocarbon turbulent premixed flames and to develop predictive models for noise that account for variations of fuel mixture and flow conditions. The high diffusivity of hydrogen molecules accelerates the burnout of reactant pockets, leading to increased direct noise emissions. Furthermore, the unique flame topologies that emerge in hydrogen and hydrocarbon flames also affect the sound produced. By utilizing data from direct numerical simulations (DNS) [1], this study will identify the key physical mechanisms and parameters influencing acoustic emissions in both types of flames. The insights gained will inform the development of hybrid models that combine data-driven approaches with physics-based principles. In this framework, data will be used to train model parameters, while physical insights will guide the definition of loss functions. Ultimately, this work aims to enhance the accuracy of large eddy simulations (LES) in predicting combustion noise from gas turbines.

Objectives:

1. Data Analysis: Conduct a comprehensive analysis of DNS data to identify the distinct characteristics of sound generation in hydrogen and hydrocarbon flames.
2. Mechanisms Identification: Investigate the underlying physical mechanisms responsible for the observed differences in acoustic emissions.
3. Model Development: Develop data-driven models for sub-grid sound emission based on the findings, tailored for integration into LES frameworks.
4. Validation: Validate the developed models against experimental data and existing theoretical models to ensure robustness and accuracy.

[1] https://doi.org/10.1080/13647830.2018.1457799

[2] https://doi.org/10.1080/00102202.2022.2041624

Academic-led project

Supervisors: Jen Muggleton and Phil Joseph

Project partner: UK Water Industry Research

Leakage from pipes is a major issue in the water, gas and oil industries, not only in environmental terms, because of wasting an important natural resource, but also in economic and health terms. In particular, UK Water Industry Research (UKWIR) have identified leakage as one of their strategic priorities and have recently set up the Zero Leakage 2050 initiative, aiming to address the question “How will we achieve zero leakagein a sustainable way by 2050 ?”. The acoustic signals generated by a leak and thence propagated along pipelines have been exploited for many decades to locate leaks but, historically, by making measurements directly on the pipes.

This exciting and innovative project aims to develop an acoustic geo-camera system using two (or more) on- or above-ground sensors to carry out synchronised multiple ground vibration measurements to locate a leak and/or a pipe system.

The project will be a balance of theoretical and experimental work, with some of the experimental work being undertaken at outdoor test sites in the UK, provided by the industrial partner; it would suit a candidate with an interest in signal processing, strong analytical/numerical skills and an enthusiasm for experimental measurements. Specifically, the following will be investigated:

• Wave propagation modelling (within the pipe and in the ground), both analytical and (possibly) numerical
• Development of signal processing methods, including beam forming techniques to combine sensor outputs
• Inverse methods to relate the surface response to the in-pipe behaviour
• Experimental measurements

Academic-led

Supervisors: Rafael Mestre, Mohammad Belal, Jen Muggleton

This PhD project aims to develop machine learning (ML) and signal processing techniques for analysing data from Distributed Acoustic Sensing (DAS) systems in urban environments. DAS technology, which uses fibre optic cables to detect acoustic signals over large distances and with extreme temporal, frequency and spatial sensitivity, offers real-time monitoring for applications such as traffic flow, pipeline integrity, and noise pollution detection. The primary challenge is efficiently processing the vast amount of data generated by DAS, especially in noisy and complex urban settings.

The student will develop novel ML models combined with classical signal processing techniques to detect and classify key events such as pipeline leaks, traffic congestion, or sound pollution using this acoustic data at different frequencies. These models will be designed to handle noisy data and operate in real-time, making them suitable for urban monitoring systems. The specific application of DAS, whether focused on traffic, infrastructure, or environmental monitoring, will remain flexible and open for the student to shape based on their research interests and the needs of the project. This project will be embedded in an ongoing grant that provides access to DAS data collected from urban environments in Southampton and London, allowing the student to test and refine the developed models in real-world scenarios. Additionally, the student will have the opportunity to engage in interdisciplinary collaboration with researchers working in fields from the social sciences to the humanities. This will provide a broader perspective on how DAS technology can be applied across various sectors, enabling the student to explore diverse applications and work in a multidisciplinary research environment.

Academic-led

Supervisors: Preethi Rajendram Soundararajan and Edward Richardson

The central theme of this research will be studying thermoacoustic instabilities of hydrogen flames. As the aviation sector envisages using hydrogen as fuel to reduce CO2 emissions, the adoption of new injection techniques may increase the susceptibility of flames to thermoacoustic instabilities. These instabilities are created due to the generation of acoustic waves from unsteady combustion, and when this happens in a resonant chamber, it further leads to heat release rate fluctuations, which may grow, making the system unstable. This phenomenon arises in high-performance devices like aircraft engines, gas turbines, and rocket thrust chambers and could pose serious problems, including intense noise and vibrations, flame flashback/blow-off, and even lead to spectacular failures. One potential injection configuration for hydrogen-based combustors is introducing high-velocity hydrogen jets in a cross-flow of air to ensure good mixing and flame stabilisation. In a preliminary work carried out by the lead supervisor (see Fig. 1), it was observed that this class of hydrogen flames exhibit transverse oscillations, resulting in alternate interactions with adjacent flames. The instability frequency corresponds to the longitudinal mode of the duct, thus posing an interesting question: how do the longitudinal flow fluctuations translate to traverse flame oscillations? This problem has not been explored well, and the proposed research will endeavour to understand it through systematic experimentation and numerical simulations. In the first step, a canonical multi-jet burner will be developed, and a range of momentum flux ratios will be studied to identify the critical conditions at which the flame becomes unstable. In the second step, numerical simulations of the canonical configuration guided by experiments will be carried out to further understand the flame dynamics. Finally, the burner configuration will be modified to study flame response to external acoustic perturbations and measure the flame transfer function to develop low-order models for instability prediction.

Academic-led project

Supervisors: Steven Bell and Stefan Bleeck

There is a large cost in terms of staff time and devices in supplying hearing aid provision.

Automation and machine learning, for example in terms of automated hearing testing, machine learning to seek information from patients and self-fitting hearing aids have the potential to reduce the staff time required to deliver services and hence to improve access to services. However it is critical to still be responsive to the needs to service users and to be safe.

The aim of this project is to explore which aspects of hearing assessment and hearing aid provision for adults in the UK might be automated. It would develop an understanding of the current care pathway in the UK, identify aspects that might be automated and then explore how that implementation could occur whilst maintaining a high-quality service for the end user. It is also important that any approaches used are considered trustworthy. The intended end point is a set of recommendations for clinical practice with some evidence demonstrating the clinical efficiency of such approaches.

Academic-led

Supervisors: Felix Langfeldt, Giacomo Squicciarini and Sjoerd van Ophem

Acoustic metamaterials emerged recently as a novel approach for controlling the propagation of sound waves. With acoustic metamaterials it became possible to engineer structures that exhibit unusual effective material properties, for example negative density. This opened up a wide range of novel ways to control sound, from extremely thin low-frequency sound absorbers to acoustic invisibility cloaks. The performance assessment of current metamaterials-based technologies is exclusively focused on objective metrics, such as absorption coefficients our sound pressure levels. What is missing in the current research landscape is a different view that looks at the human perception of how sound is altered by acoustic metamaterials and if this can be exploited to make noise sound more pleasant, without necessarily improving objective metrics.

The main aim of this project is to create virtual models that can auralize the sound of acoustic metamaterials in different applications, for example as noise treatments or in musical instruments. These models can be based on established numerical methods (e.g. reduced order finite element models) or novel analytical approaches that can be developed during the project. A key property of the models should be the ability to change the properties of the acoustic metamaterial treatments (e.g. their tuning or the type of metamaterial) to be able to study the sound-changing effects of the metamaterials. To validate the virtual auralization models, the world-class acoustic testing facilities at the Institute of Sound & Vibration Research can be used. Finally, the virtual models will be used in listening tests to measure the effect of metamaterials on the subjective perception of sounds by humans and contrast this to the established objective measures. The novel techniques developed in this project can be used to let the general public listen to the effect of AMM on these sounds through an interactive web app.

Project type: partner-led

Supervisors: Bahareh Zaghari and Giacomo Squicciarini

Project partner: iNetic

Addressing the noise and vibration challenges posed by new sound sources in electrified drivelines is a fundamental step in the sustainability of transportation. By minimising noise and vibration while optimising weight, size, and maintenance costs, one can create quieter and more efficient vehicles. This not only improves user comfort but also reduces carbon and non-carbon emissions and mitigates environmental noise pollution. These advancements make cleaner transportation more affordable and accessible to a broader segment of the population, contributing to a more sustainable future.

In an integrated driveline system comprising an energy source, electrical machine, inverter, and gearbox, noise and vibration can originate from various sources, including electromagnetic forces, mechanical imbalances, and high-frequency switching. These components, when coupled together, can amplify vibrations and noise. For example, electromagnetic noise from the motor can interact with gear meshing vibrations, while high-frequency switching noise from the inverter can propagate through the structure, causing complex vibrational patterns. The integration challenges include managing structural resonances, mitigating torque ripple, and ensuring that vibrations from one component do not negatively impact the performance or safety of others.

To effectively mitigate noise and vibration in such systems, the student will develop a holistic modelling approach that balances total weight and size of the system, performance and efficiency, with noise control strategies. This includes vibration isolation sound insulation, and changes to the design of the parts.  The student will also characterise the mechanical and electrical behaviour of the different components in isolation and when coupled to a  host structure and develop algorithms to evaluate and mitigate the vibration response. After the modelling stage, the selected designs of fully integrated driveline, will be built and tested at iNetic.

Academic led

Supervisors: Michal Kalkowski and Tiina Roose

Sound plays a role in plant ecology, but the understanding of its significance is still in its infancy. Recent results suggest that plants emit sounds indicating their condition and react to sounds in their vicinity (e.g. pollinators). However, most studies on the topic are mere observations focusing on biological processes. This project aspires to advance plant acoustics by offering a solid physics-based foundation for our understanding f plants’ sensory and transmission capabilities. The project involves developing vibration/wave propagation models to understand plant dynamics in response to vibroacoustic stimuli across broad frequency ranges. It builds on recent biological research suggesting plants’ biochemical response to incoming signals, and acoustic emissions. The candidate will integrate theoretical modelling with measurements to analyse plant behaviour. They will also work on determining fundamental acoustic and vibration models to identify features responsible for receiving signals with given properties (e.g. frequency bandwidth). These efforts will be supplemented by the analysis of wave propagation in plant material and coupling with the surrounding medium to study energy transfer.

This project hopes to find applications in environmental monitoring, agriculture and biological sciences. This interdisciplinary study requires a candidate with a background in engineering, applied physics, acoustics, or a related field. The role includes model development, vibration/wave measurements, and collaboration with experts in related disciplines. The project offers the opportunity to contribute to a novel area of research, bridging physics and biology.

Academic-led project

Supervisor: Matthew Wright

Project partner: The OHMI Trust, a charity that supports disabled musicians

Being able to play a musical instrument is an important gateway to artistic and personal self-expression, and to participation in social activities. Upper-limb differences and disabilities can present a barrier to all these desirable outcomes. Organisations such as OHMI do valuable work to develop solutions such as adapted instruments, but these are often expensive because they are produced in small quantities. The most successful accessible instrument would appeal to all players, disabled or not.

Most wind instruments work by the player covering fingerholes, pressing valves etc, to change the input acoustic admittance seen by the mouthpiece. The peaks and troughs in its admittance function determine the frequency that the mouthpiece (fipple, reed, lip-mouthpiece combination etc) oscillates at and hence the note that the instrument plays.

This project is to develop a class of programmable-impedance wind Instruments, which would use active control of the waves in the air column to control its input impedance via electroacoustic actuators and sensors in the pipe. This could be attached to a regular mouthpiece and blown expressively as normal, but the note-selection process would be DSP-controlled and the input from the player could come via a range of possible interfaces: MIDI keyboard, touch-sensitive pad, etc.

This would allow a disabled musician who can’t, for instance, cover fingerholes to play such an instrument, while retaining expressiveness (unlike conventional digital wind instruments). It would also allow musicians to extend the capabilities of their instrument by allowing effects such as vibrato, by dynamically changing the impedance curve to affect the pitch or tone of the note, meaning its appeal would be wider than just those who need accessible instruments.