UTTOP–LGP PhD: Physics-Informed Real-Time Surrogate Modelling for Linear Friction Welding

UTTOP–LGP PhD: Physics-Informed Real-Time Surrogate Modelling for Linear Friction Welding

Research question and project fit

Can a model predict linear friction welding quickly enough to be useful during a weld, while still representing material flow, bonding and defects? This PhD combines equations and multiscale measurements to develop a computational component for a future digital twin. It suits thermomechanics, finite-element and Python experience with an interest in scientific machine learning. The research must establish both physical reliability and prediction speed; delivery of a complete industrial twin or closed-loop controller is outside the advertised scope.

The employer-authored notice published in October 2026 identifies Université de Technologie Tarbes Occitanie Pyrénées (UTTOP), Laboratoire Génie de Production (LGP), and Tarbes, France. Its publication date is 6 October 2026, deadline 23 November 2026, and planned start 1 September 2027. It declares EU funding but gives no remuneration, funded duration or grant name. The named supervisor and a publicly usable project contact are also unresolved. Treat this as worth a targeted enquiry and preparation, conditional on eligibility and contract clarification, rather than an unconditional fully specified offer. Recruitment notice.

The principal strength is relevant experimental and numerical welding expertise in the host environment. The principal risk is that the surrogate must learn a strongly coupled transient process with moving interfaces from potentially limited data. For a Chinese applicant, general French degree and visa routes exist, but no inspected project source explicitly confirms Chinese-national eligibility under this funding arrangement. Obtain that answer before investing in a substantial application.

Research problem and probable doctoral work

LFW joins materials in the solid state through relative oscillation and pressure. Friction and plastic deformation create heat; changing temperature then affects constitutive response, friction, flow and bonding. Macroscopic variables such as temperature, stress, force and flash formation interact with local microstructural evolution. A model that predicts temperatures accurately but misses insufficient bonding would not solve the advertised problem.

The proposed approach combines surrogate modelling, particularly physics-informed neural networks (PINNs), with multiscale experimental information. Equations and boundary constraints are intended to improve physical consistency, while measurements and numerical simulations provide complementary information. The recruitment description explicitly limits the project to the computational component of a future twin. It does not include delivery of a complete digital twin or its closed-loop control and decision architecture.

An appropriate work programme would first define the material, geometry, process stages and observable quantities; build or reproduce a thermomechanical reference model; then establish an experimentally checked dataset and a simpler surrogate baseline. Only after that should the doctoral researcher test which physical constraints improve generalisation. These are analytical recommendations, not an advertised work-package schedule. They keep the engineering question visible: what accuracy and latency are necessary for an estimate to be useful during a weld?

The critical tests should separate interpolation from extrapolation to new operating settings or materials. Training, mesh refinement and experimental calibration occur offline, whereas inference latency must be measured on the intended hardware. Report both costs. Defect prediction should use independent weld-level validation, with clear labels and uncertainty; randomly splitting highly correlated samples from one run could overstate transfer. Conservation residuals, force and temperature errors, spatial field agreement, and false-negative rates address different failure modes. No source inspected supplies a project-specific latency target, dataset size or validated LFW benchmark.

There is sound methodological reason to question the assumption that adding physics automatically makes training reliable. Krishnapriyan, Gholami, Zhe, Kirby and Mahoney demonstrated difficult optimisation in PINN benchmarks involving convection, reaction and diffusion, and investigated curriculum and sequential learning. This is external methodological evidence, not LGP output or a welding validation. It supports testing conditioning and training strategy, while leaving moving-contact mechanics unresolved. Original NeurIPS paper.

Host laboratory, supervision and recent papers

LGP's mechanics/materials department provides a relevant setting: its M²SD group works on microstructure, constitutive identification and thermomechanical simulation, while TM²P covers tool/material interactions, tribology, FSW and in-operando characterisation. This is evidence of related capability. It does not establish that an LFW machine, GPU allocation or a particular industrial dataset is reserved for this doctoral researcher. Department research activities, LGP official site.

No named supervisor is publicly assigned in the inspected advert. Olivier Pantalé's official profile describes nonlinear constitutive modelling and FE software development; Vincent Wagner has an official LGP profile and appears in the welding papers below. These establish potential surrounding expertise, not their appointment to this project. It would be misleading to present either person as the confirmed thesis director. Ask who will supervise the mechanics, machine-learning and experimental components, and whether they already collaborate routinely. Pantalé profile, Wagner profile.

Three important recent original papers clarify what the surrounding research actually contributes:

Ambrosio and colleagues (2022): peak-temperature estimation

Ambrosio et al., 2022: semi-empirical peak-temperature estimation in FSW. The authors combine thermocouple measurements in three aluminium alloys with a polynomial temperature model and X-ray weld-quality observations. This is a useful reduced-model and experimental-validation precedent, but predicts peak temperature rather than complete LFW fields. Ambrosio is first author; Wagner, Dessein and Garnier are LGP coauthors. No individual contribution statement was inspected for this paper, so their personal methodological roles are not inferred. Publisher abstract and affiliations.

Ambrosio, Tongne, Wagner, Dessein and Cahuc (2023): damage-aware flow

Ambrosio, Tongne, Wagner, Dessein and Cahuc, 2023: damage-aware material-flow prediction in FSW. Their three-dimensional coupled Eulerian–Lagrangian model incorporates damage mechanisms informed by orthogonal cutting and examines mesh sensitivity. It provides a precedent for checking whether a numerical reference reproduces flow, rather than accepting temperature agreement alone. The public CRediT statement assigns methodology and validation to Ambrosio, methodology to Tongne, analysis/investigation to Wagner, and supervision/resources to Dessein. These roles concern this paper; they do not appoint the current supervisor. FSW tool-driven flow also differs from LFW's oscillating interface. Publisher preview, methods and contribution statement.

Ambrosio and colleagues (2022): acoustic observations

Ambrosio, Dessein, Wagner, Yahiaoui, Paris, Fazzini and Cahuc, 2022: acoustic emission in FSW. Static tool tests and sensor-location comparisons precede weld experiments; changes in signal features are associated with material behaviour and internal defects. This supports thinking carefully about observability and sensor placement. It does not demonstrate a universal defect detector, causal microstructure reconstruction or LFW transfer. Ambrosio is first author and the remaining named authors are coauthors; the inspected institutional record does not allocate individual tasks. Institutional publication record and abstract.

These are host-related publications from 2022–2023, not papers attributed to an unnamed supervisor. The reading scope comprises abstracts, metadata and publicly accessible previews, with the material-flow paper also exposing introduction, methods snippets and CRediT. Full subscription articles were not accessed. The lab's record of Ambrosio's 2021 FSW doctorate and its official publication directory provide additional evidence of continuity, without proving the current grant is the historical ENABLE project. Completed theses, Publication access.

University, doctoral programme and field outlook

UTTOP is a public technology university built around ENIT and the Tarbes IUT, with engineering and technology education extending to doctoral research. Its institutional signal is a specialised engineering environment. A verified separate global numerical rank was not established; a Toulouse umbrella rank would not constitute an individual UTTOP or LGP rank. Research fit, facilities and actual supervision are more useful here than assigning an unsupported league-table position. Institutional presentation.

Working at LGP and receiving the degree are separate matters. The current lab page says most doctoral students register at Toulouse INP and lists several schools in Toulouse, plus SEA in Pau. MEGeP is thematically plausible, but this offer does not identify its school or degree-awarding institution. Request the exact registration route for the 2027 intake, training requirements and supervision agreement. French national rules provide for admission and annual review through the institution and school; the normal full-time research duration is a framework, not proof of this vacancy's paid term. LGP doctoral route, Schools listed by LGP, National doctoral framework.

Career implications

The skills can transfer to process simulation, manufacturing R&D, scientific software and experimentally validated scientific machine learning. This is an analytical career assessment based on the work, not evidence of a measured vacancy boom or guaranteed aerospace placement. A strong thesis would make its constitutive assumptions, validation split and computational costs reproducible. Industrial data restrictions could complicate publication and portfolio building, so ask about access, intellectual property and publishable outputs early. The national doctoral framework recognises both public and private career pathways, but project-specific graduate outcomes were not identified.

Funding, fees and living budget

Only EU funding is confirmed for the offer. Do not infer MSCA status, mobility allowances, family allowances, full tuition reimbursement or a three-year employment contract from that label. The public doctoral-contract framework provides a €2,300 gross monthly research-only minimum from January 2026 and a usual initial three-year term if that legal contract is used. This is a conditional benchmark, not the project's confirmed package, and it is not net take-home income. Ask for the employer, contract type, gross salary, deductions, teaching obligations and funding end date. Government doctoral funding guidance.

Current public references put ordinary national doctorate registration at €398 and CVEC at €105 for 2026–2027. The advertised start falls in 2027–2028: its applicable rates, exemptions and reimbursement are unknown. A funded employee is not automatically a means-tested scholarship holder. Confirm the bill and who pays it instead of describing the position as fee-free. Public registration fees, CVEC official FAQ.

UTTOP's July 2024 accommodation page quotes studios around €286–375 and shared accommodation around €300–360 monthly, with a one-month-rent deposit. These are dated examples subject to availability, not 2027 offers. For planning, an applicant-created scenario might allocate €450 housing/utilities, €250 food, €100 local transport/phone and €200 personal/health/contingency, totalling €1,000 monthly before fees, travel and major setup costs. This illustrative sum is not an official Tarbes cost survey. Refresh actual rents and utilities before comparing it with confirmed net pay. Dated accommodation information.

Budget without assumed APL housing assistance. UTTOP's page warns of restrictions for non-EU non-scholarship students from July 2026. Service Public explicitly notes possible eligibility for students undertaking professional employment, while CAF specifies the student-residence context. Thus neither automatic assistance nor categorical exclusion of a salaried Chinese doctoral researcher follows. The actual permit, employment, resources and bursary status need assessment. University guidance, Government reform and employment exception, CAF current explanation.

The university explicitly provides Welcome Desk support for LGP doctoral students, with Eliane Casta listed for that profile. This is an administrative housing contact, not the advertised research supervisor. Campus facilities include housing, catering and transport connections; they do not guarantee a room. Doctoral Welcome Desk contact, Campus practical information.

Applicant fit and preparation

The advert targets a master's or engineering graduate in mechanics, scientific computing or materials science, with finite-element modelling, thermomechanics and Python. Abaqus and AI/PINN experience are desirable; it states upper-intermediate English. No inspected text requires a named test score or expressly waives French. An applicant should ask which documents establish language proficiency and what languages are used in training and laboratory work.

A mechanics applicant is well placed if they can explain contact, constitutive calibration and experimental error, then demonstrate a small physically constrained learning example. An AI applicant needs credible mechanics preparation; a generic image classifier is weaker evidence than a validated transient heat or elastoplastic problem. A materials applicant should show how microstructure measurements become model targets or constraints, rather than assuming every observed feature is identifiable from machine signals.

A practical portfolio could contain one reproducible FE sensitivity study, one Python surrogate compared with a simple baseline, and a short experimental-validation discussion. These are recommendations, not mandatory institutional documents. Explain where the model fails and how held-out process settings differ from training data. For the motivation letter, tie previous work to the particular bottleneck—contact, transient heat, flow or reliable defect labels—and identify the missing skill you intend to learn.

For Chinese applicants, national rules allow assessment of master-level or equivalent preparation, and official immigration guidance describes a researcher route requiring a master-equivalent qualification and a hosting agreement. Neither establishes this vacancy's nationality restrictions, grant mobility rules or security/access clearance. Chinese eligibility remains unverified. Request written confirmation that the applicant's nationality and degree can be considered, and which equivalence documents and residence route apply. Researcher visa framework.

Questions before committing

Before applying: Who is the named thesis director and the correct recipient/submission route for ABG140423? Are Chinese nationals and a Chinese master's eligible under the exact grant and facility-access rules? Is the 1 September 2027 start correct, and what time zone and submission time apply to the 23 November deadline? The employer vacancy index does not currently list this exact offer; that omission is not proof of closure, but makes direct process clarification useful. Current employer index.

At interview: What material, machine and dataset already exist, and what new experiments must the student develop? Who supervises mechanics, PINNs and microscopy, with what meeting frequency? Which accuracy, inference latency and out-of-distribution tests define success? Is a simpler reduced-order model an acceptable outcome if PINNs underperform? Which data and results can be published?

Before accepting: Obtain the signed-term outline covering gross remuneration, paid duration, duties, fees and any funding gap. Confirm the degree-awarding institution and school for 2027, international degree assessment, hosting agreement and arrival support. Request actual housing options and a net-pay estimate; assess allowances from the applicant's real status. These unresolved details affect feasibility, but do not justify dismissing a scientifically substantive project merely because the university is specialised.

Evidence and Sources

Evidence was checked on 12 October 2026; the employer notice is dated 6 October. The three host-related studies retain their original abstract/metadata/preview limits, with methods snippets and CRediT available for the flow study; full subscription texts were not accessed. The external PINN benchmark remains separate from LGP output or LFW validation. These dates do not establish current vacancy status, 2027 costs or personal nationality/contract eligibility.

• Recruitment notice ABG 140423
• UTTOP institutional identity
• LGP official site
• LGP mechanics-materials-processes department
• Olivier Pantale official staff profile
• Vincent Wagner official staff profile
• Doing a doctorate at LGP
• LGP doctoral-school list
• LGP completed theses
• LGP publication access
• Semi-empirical FSW peak-temperature model
• CEL damage and material-flow prediction in FSW
• Acoustic emission in FSW
• Characterizing PINN failure modes
• French doctoral funding framework
• National doctoral training framework, consolidated
• Researcher immigration route
• Public higher-education registration costs
• CVEC official FAQ
• UTTOP accommodation guide
• Welcome Desk registration and doctoral support
• UTTOP current housing-allowance guidance
• Current Tarbes campus practical guide
• Current employer research-vacancy index
• July 2026 housing-allowance reform and employment exception
• CAF 2026 student guidance