University of Turku Digital Manufacturing Master’s Thesis Guide: KTEK0015, 30 ECTS, Additive Manufacturing, ML and UTUGradu

What this guide covers

This guide is for the University of Turku Digital Manufacturing specialisation track in the current 2024-2027 Mechanical Engineering curriculum. It is anchored to the exact Peppi object KTEKMSCDIGMAN2427, programme 99992, rather than a generic manufacturing thesis model. The central structural rule is that the 40 ECTS Master’s Thesis and Project category contains four separate academic objects: KTEK0015 thesis 30 ECTS, KTEK0026 thesis seminar 5 ECTS, TTDK1308 maturity examination 0 ECTS, and KTEK0016 Project Work 5 ECTS. The guide then shows how additive manufacturing, materials processing, sensor data, machine learning, joining, surface engineering, virtual manufacturing and industrial evidence should be bounded in a defensible thesis.

Current programme object

The current Peppi programme object is KTEKMSCDIGMAN2427, programme 99992. The degree is Master of Science (Technology), formally 120 ECTS and two years, organised through the Faculty of Technology and the Department of Mechanical and Materials Engineering. Digital Manufacturing is one of the Mechanical Engineering specialisation tracks. Use the curriculum period attached to your own study right and HOPS when planning the thesis. Even where a neighbouring Mechanical Engineering track uses the same thesis-course codes, its track-specific methods and evidence expectations should not be copied automatically into Digital Manufacturing.

How the 120 ECTS degree is structured

The current programme combines 20 ECTS Joint studies in Mechanical Engineering, 20 ECTS Digital Manufacturing studies, a 40 ECTS Master’s Thesis and Project category, 20-25 ECTS Minor or Thematic Studies and 15-20 ECTS Other Studies. Peppi models the Advanced Studies component as exactly 80 ECTS. The 115-125 ECTS range at the programme root reflects selectable ranges in the surrounding modules and should not be presented as a different formal degree size. For thesis planning, the distinction between the 20 ECTS track studies and the 40 ECTS thesis-and-project category matters because they support different learning and assessment purposes.

Read the 40 ECTS Master’s Thesis and Project category correctly

The 40 ECTS category is not a 40 ECTS thesis. KTEK0015 is the individually examined 30 ECTS master’s thesis. KTEK0026 is a separate 5 ECTS thesis seminar. TTDK1308 is a 0 ECTS maturity examination. KTEK0016 is a separate 5 ECTS Project Work course. All four sit inside the same category, but none should be silently merged into KTEK0015. If project work or seminar activity contributes to the same industrial problem, distinguish the assessed outputs, new individual research contribution and evidence generated specifically for the thesis.

Exact thesis course: KTEK0015

KTEK0015 Master’s Thesis in Technology, Mechanical Engineering is the exact 30 ECTS thesis course. Its public course evidence frames the thesis as scientific work requiring appropriate research methods, scientific literature, field knowledge and scientific writing. In Digital Manufacturing, a thesis can involve experimental process development, materials characterization, sensor-based monitoring, ML, simulation, joining, coating or an integrated industrial problem. The engineering artefact is not the thesis by itself. A defensible manuscript connects a specific research question to a reproducible method, transparent evidence and a conclusion that stays within the tested scope.

The separate KTEK0026 thesis seminar

KTEK0026 Master’s Thesis in Technology Seminar, Mechanical Engineering is a separate 5 ECTS course inside the 40 ECTS category. The seminar supports research planning, theory and literature use, methodological choices and presentation of the research plan and later results. Its credit remains separate from KTEK0015. Use seminar feedback to improve the thesis question, study design and communication, but do not describe seminar participation as thesis credit or use completion of the seminar as proof that the final research method is valid.

The separate KTEK0016 Project Work course

KTEK0016 Project Work, Mechanical Engineering is a separate 5 ECTS course. It may include an accepted project plan, industrial work, project journal and final report. A project can generate a machine setup, preliminary dataset, prototype, company context or process concept that later helps the thesis, but project delivery does not replace individual thesis research and examination. If the same company problem spans KTEK0016 and KTEK0015, agree early which work belongs to the project course and what new research question, analysis and evidence belong to the thesis.

TTDK1308 maturity examination

TTDK1308 is the 0 ECTS Degree Qualifying Examination for the master’s degree. The normal route can use the thesis abstract or another suitable thesis component to demonstrate familiarity with the thesis field, while a conditional written route can apply depending on previous degree and Finnish or Swedish educational-language background. The route is student-specific. Do not infer your own requirement from another student’s experience. Recheck the current HOPS and University instructions before final submission so the thesis process is not delayed by an unresolved maturity requirement.

Finding a topic and supervisor

Digital Manufacturing topics can arise from University research, companies or wider industrial development. A strong topic is narrower than “improve additive manufacturing” or “apply AI to production.” Define the process, material, machine, product or decision being studied and the evidence needed to answer one tractable question. For example, a thesis could test whether a specified sensor representation predicts a defined defect under a bounded process window, or whether a post-processing treatment changes a measured material property under controlled conditions. The supervisor should help keep the question realistic for 30 ECTS.

Build a thesis plan around the manufacturing evidence chain

Before the main experiments, define the research question, process boundary, material or feedstock, machine configuration, input parameters, sensor channels, specimen or build plan, quality reference, analysis rules, validation strategy, permissions, risks and milestones. Digital Manufacturing projects often contain long dependencies: machine booking, feedstock availability, repeated builds, destructive testing, external microscopy, industrial data access or model training. A written plan makes these dependencies visible and helps prevent the research question from changing after favourable results appear. Predefine important exclusions and quality metrics where possible.

Use supervision as evidence control

Use supervision to review methodological decisions, not merely report progress. Bring failed builds, changed parameter windows, sensor dropouts, relabelled defects, excluded specimens, post-processing changes, unexpected microstructures, model revisions and company constraints to supervision. Record major decisions so the Methods section can explain what actually happened. If the project moves from process monitoring to quality prediction or from a laboratory coupon to an industrial component claim, recheck whether the evidence supports the expanded level. Scope changes should be deliberate rather than hidden inside the final Discussion.

How examination and grading work

The current KTEK0015 evidence states that at least two examiners evaluate the thesis using University evaluation and grading guidance. Final acceptance is decided by the head of department, with the grade based on examiner evaluation. Company satisfaction, successful machine operation or supervisor support therefore does not substitute for academic examination. Write the thesis so an examiner can trace the question through literature, method, process inputs, raw or derived evidence, uncertainty, validation, results and conclusion. A technologically impressive demonstration cannot compensate for missing controls or unsupported generalisation.

Choose the correct evidence level

Digital Manufacturing spans several evidence levels: feedstock or raw material, process parameters, in-process sensor signal, manufactured specimen, microstructure, material property, component quality, production-system performance, economic feasibility and lifecycle effect. These levels influence one another but are not interchangeable. A cleaner sensor signature does not automatically prove a stronger part. Lower porosity does not automatically prove longer service life. A successful laboratory print does not establish production capability. State the evidence level in the research question and only move upward when additional measurements or validated models justify the transition.

Organise the literature around process, material, data and quality

A useful literature review should separate the manufacturing process, material state, sensing or data method, physical quality measurement, modelling method and industrial application. Compare published results only when process family, material, geometry, parameter range and quality definition are sufficiently compatible. Avoid ranking processes from isolated headline values. When a paper reports ML performance, ask how the labels were generated and whether builds or machines overlap across data splits. When it reports material performance, check the specimen state and post-processing. The review should expose where evidence is comparable and where it is not.

Digital Factory: define what the digital representation proves

KTEK0011 Digital Factory covers digital modelling of manufacturing operations, digital twins, manufacturing-data collection and processing, Industry 4.0, industrial IoT, cloud data, PLM and MES. A digital representation can support traceability or decision-making, but it does not automatically prove physical product quality. If the thesis uses a digital twin, define what physical system it represents, how it is updated, which variables are synchronized and what validation was performed. If the claim concerns productivity, quality or resource use, measure that outcome rather than treating digital connectivity as the outcome itself.

Additive manufacturing: report the complete process context

KTEK0012 and KTEK0063 make additive manufacturing a central Digital Manufacturing method. A thesis should identify the process family, machine, feedstock or material condition, build orientation, key parameters, geometry, support strategy and relevant post-processing. Report enough context to make a performance comparison meaningful. Dimensional accuracy, surface finish, porosity, microstructure and mechanical properties are separate quality dimensions. A part that prints successfully under one setting shows feasibility under that setting, not repeatable industrial process capability. Preserve failed or excluded builds when they affect interpretation.

Separate feedstock, process and post-processing effects

Manufactured properties can depend on feedstock characteristics, process parameters, thermal history and post-processing. If several of these change together, the thesis cannot attribute the final difference to one factor without a suitable design. Characterize the feedstock when particle size, morphology, moisture, composition, rheology or another property can influence the process. Separate as-built and heat-treated states. If hot isostatic pressing, machining, surface finishing or another downstream operation changes the result, report it explicitly instead of attributing the final property entirely to the printing process.

Materials engineering: connect process, microstructure and property carefully

KTEK0034 emphasizes material classes, solidification, phase transformation, powder metallurgy, sintering, microstructure, feedstock and heat treatment. A Digital Manufacturing thesis may legitimately connect process to microstructure and microstructure to property, but each link needs evidence. One representative micrograph is not population-level proof. A hardness change does not automatically prove a specific phase mechanism. If anisotropy or build direction matters, report specimen orientation and test direction. Distinguish observed association from demonstrated mechanism, especially when the mechanism is inferred from literature rather than directly measured.

Materials processing technologies: compare like with like

KTEK0033 includes welding, laser-based processes, thermal spray and other manufacturing processes, including their advantages and limitations. Process comparison should use application-relevant criteria such as material compatibility, geometry, heat input, quality, productivity, inspection requirements and post-processing. One process being faster under one condition does not make it universally superior. Define the baseline and operating window. If sustainability or cost is discussed, state the system boundary and assumptions rather than inferring broader benefit from one technical metric such as deposition rate or material utilisation.

Machine learning in Digital Manufacturing: define the target first

KTEK0070 focuses on ML for manufacturing process monitoring, sensorisation, data acquisition, defect or anomaly detection and quality assurance. Start by defining the target: what exactly is predicted or classified, and what reference establishes the truth? A vague “quality” label is not enough. Explain sensor channels, sampling, preprocessing, feature construction and label provenance. If the target is a defect, define the defect and inspection method. If the target is a material property, explain how that property was measured. The model should answer a manufacturing question, not merely achieve a high score.

Prevent data leakage across builds, parts and machines

Manufacturing data often contain strong dependence within the same build, part, batch, machine or time window. Randomly splitting individual rows can leak near-duplicate process states into both training and test sets, producing an unrealistically high score. Group the split at the level that matches the generalisation claim. If the claim is performance on unseen builds, hold out whole builds. If it is transfer to another machine, test on another machine. Keep preprocessing fitted only on training data when appropriate, and document how repeated measurements and temporal dependence were handled.

Choose ML metrics that match the manufacturing decision

Accuracy alone can be misleading when defects are rare. Report metrics that reflect the operational decision, such as precision, recall, specificity, F1, ROC/PR measures, calibration or regression error, depending on the task. Interpret false positives and false negatives in manufacturing terms. A false negative may allow a defective part to pass; a false positive may cause unnecessary scrap or inspection. State the threshold used and how it was selected. A threshold chosen after seeing the test set should not be presented as independent validation.

Offline ML performance is not real-time process control

A model can perform well offline and still fail in production because sensor drift, latency, machine change, different material lots or unobserved process states alter the data distribution. Distinguish detection from prediction, and prediction from closed-loop control. If the thesis only evaluates recorded data, do not claim real-time control performance. If online deployment is studied, report inference timing, missing-data behaviour, alarm logic and how the model interacts with the process. A high offline score does not establish safe or effective automatic control.

Do not turn ML correlation into a physical mechanism

ML can identify predictive patterns without explaining the physical mechanism. Feature importance, attention weights or SHAP-style explanations may help interpret a model, but they are not automatically causal evidence. If the thesis claims that a process variable causes a defect mechanism, support that claim with process physics, controlled variation, material characterization or other appropriate evidence. Keep three claims separate: the variable predicts the outcome, the variable is associated with the outcome, and the variable causes the outcome. The evidence required becomes stronger at each step.

Digital Joining: define the joint and quality reference

KTEK0062 covers joining processes, quality procedures, mathematical modelling, process monitoring and digital quality systems. A joining thesis should identify materials, joint geometry, preparation, process settings and the quality criterion. Sensor signals, thermal traces or camera outputs are not automatically equivalent to destructive testing, metallography or other reference-quality measurements. If a monitoring threshold is proposed, validate it against suitable reference evidence. State whether the goal is detecting a process anomaly, predicting a defect or certifying a joint, because those are different evidence claims.

Joining models need physical and measurement validation

A welding or joining model is conditional on heat-source representation, material properties, contact or boundary conditions and numerical assumptions. A numerically stable model can still be physically wrong. If temperature, deformation or residual stress is important, compare with suitable measurements or trusted benchmarks where feasible. If the model is calibrated using one dataset, distinguish calibration from independent validation. Do not present a fitted model as a universally valid digital twin. State the joint type and process range over which the evidence was actually obtained.

Surface and coating studies: define degradation and service conditions

KTEK0035 covers thin and thick coatings, mechanical and thermal surface engineering, and degradation mechanisms such as wear and corrosion. Surface performance should be tied to a defined substrate, coating process, thickness or condition, environment and test method. Wear resistance and corrosion resistance are different properties. One accelerated laboratory test does not automatically establish service lifetime. If a coating performs better under a selected test, state that result at the test level unless the translation to real service conditions has been validated.

Advanced additive manufacturing: separate optimisation from qualification

KTEK0063 includes advanced AM design, material properties, post-processing, simulation, optimisation, process control, cost and business potential. This breadth creates a common thesis risk: an optimised parameter set is described as a qualified industrial process. Keep optimisation, repeatability, component qualification and production readiness separate. An optimum is conditional on the chosen objective and parameter range. Repeatability needs repeated builds. Qualification may need standards, inspection and performance evidence. Production readiness adds machine capability, throughput, maintenance, quality assurance and economic constraints that a laboratory experiment may not test.

Bio-based additive manufacturing: bio-based does not automatically mean sustainable

KTEK0072 covers additive manufacturing with bio-based materials and biopolymers, including process principles, material selection, rheology and manufacturability. Printability is not the same as final application performance. Bio-based origin is not by itself proof of lower lifecycle impact. If sustainability is discussed, define the comparison, lifecycle stages, material source, energy use, durability and end-of-life assumptions relevant to the claim. If AI is used to propose material formulations, the proposal still needs appropriate experimental or computational validation before performance conclusions are made.

Virtual Manufacturing: keep virtual and physical evidence distinct

KTEK0032 Virtual Manufacturing is a recommended study and covers manufacturing simulations, virtual commissioning, virtual factories, VR/AR and 3D scanning. A virtual model should clearly identify the real process or system it represents. Simulation outputs are not measured shop-floor results. If the thesis claims productivity, ergonomic or commissioning benefits, define the comparison and measurement used. The current course note about availability in a specific academic year should not be generalised to later years without checking current Peppi information. Course availability and thesis-valid research methods are separate questions.

Combine simulation and experiment without mixing evidence

A strong Digital Manufacturing thesis may combine thermal or process simulation, sensor monitoring, microscopy, mechanical testing and ML. Keep each evidence stream explicit. Simulation predicts under model assumptions. Sensors measure selected process signals. Characterization observes physical features. Mechanical tests measure performance under defined conditions. ML predicts a labelled target from data. Explain how these streams connect and where they remain independent. If experimental data calibrate a model, do not reuse the same observations as independent validation without acknowledging the dependency. Label measured, simulated, predicted and inferred results visibly.

Use uncertainty and replication where they change the conclusion

Manufacturing results can vary between specimens, builds, machines and material lots. Replication should match the claim. Repeated measurements on one specimen do not automatically establish build-to-build repeatability. If uncertainty can change whether a parameter setting is ranked first, whether a part meets a limit or whether a model is considered accurate, report it. Distinguish measurement uncertainty, process variation and model uncertainty where relevant. Do not report excessive decimal precision when the underlying process or instrument does not support it. Explain the sampling unit used in statistical analysis.

Company work and NDA boundaries

Digital Manufacturing theses often use industrial machines, proprietary parameter windows, production data or component drawings. Resolve confidentiality before embedding sensitive data in the manuscript. Agree what can be published, what can be generalized or anonymised and how examiners can access the evidence needed for academic assessment. Company approval of a deliverable is not thesis acceptance. An NDA should not make the scientific method impossible to examine. If exact settings cannot be disclosed publicly, explain the methodological consequence and work with supervisors on an examinable representation.

Keep manufacturing data and analysis reproducible

Preserve the provenance needed to reconstruct the result: material lot, feedstock state, machine identity, parameter files, build identifiers, sensor configuration, calibration, raw signals, preprocessing scripts, defect labels, inspection records, specimen preparation, characterization settings, model versions and final figure sources. Record important deviations from the planned procedure. A final dashboard or chart is not enough if the route from raw evidence to conclusion cannot be reconstructed. Confidential industrial files need not be public, but the research process should remain auditable to supervisors and examiners.

AI use must remain accountable

University guidance on responsible AI does not transfer responsibility away from the student. If generative AI assists with code, literature triage, translation, data cleaning or drafting, verify the output and follow current University, faculty, course and supervisor instructions. Never allow a tool to invent references, process parameters, sensor values, material properties or manufacturing results. Distinguish generative AI used as a productivity aid from ML or AI that is itself the research method. If an AI model influences a manufacturing decision, document the scientific validation appropriate to that role.

Make figures and tables auditable

Every major figure should identify what kind of evidence it represents. A process plot should state machine, material and parameter context. A sensor plot should identify channel, units, sampling and preprocessing. A micrograph should include scale and preparation information. A confusion matrix should identify the test set and label definition. A simulated field should state model conditions. If data are normalized or aggregated, explain how. Readers should be able to trace a headline result to the relevant build, dataset, specimen or model rather than treating a polished figure as self-validating evidence.

A defensible thesis chapter structure

The current public evidence does not impose one universal chapter template for every Digital Manufacturing thesis. Use a structure that makes the evidence chain easy to audit. A practical pattern is Introduction; Background and Literature Review; Research Questions; Materials, Data and Methods; Results; Discussion; Conclusions; References; and appendices where needed. A process-development thesis may add a Manufacturing Process section. An ML thesis may separate data, model development and validation. Keep Results focused on observed or calculated evidence and Discussion focused on interpretation, uncertainty, limitations, industrial relevance and bounded implications.

Turnitin checks originality, not manufacturing validity

University of Turku degree theses use Turnitin as part of originality checking. Similarity review matters for source use and academic integrity, but low similarity does not validate an ML model, a process window, a microstructure interpretation, a joining-quality claim or an additive-manufacturing parameter set. Technical validity comes from appropriate methods, controls, traceable evidence, uncertainty treatment, validation and examiner review. Treat Turnitin as an integrity control, not a technical quality certificate. Automated AI-related indicators should likewise not replace evidence-based academic judgement.

UTUGradu submission, examination and publication

UTUGradu manages the electronic higher-degree thesis process, including originality checking, examination, approval, publication and archiving. Operational details can change independently of the stable KTEK0015 credit rule, so recheck current instructions when submitting. Confirm that the intended final manuscript is uploaded, metadata are correct, confidential information has been handled appropriately, and the maturity route relevant to your background is resolved. Keep the final archived manuscript consistent with the version evaluated by examiners, and do not rely on an old cohort’s screenshots when current University instructions are available.

A practical Digital Manufacturing thesis timeline

A workable sequence is: confirm the current HOPS and 40 ECTS package; identify topic and supervisor; define the process and evidence level; resolve company access and confidentiality; create the thesis plan; run a small feasibility build or data audit; lock the main parameter and sampling strategy; collect the primary evidence; perform characterization or quality reference measurements; train or simulate only after the data path is understood; validate; analyse uncertainty; draft Methods and Results early; complete seminar and project obligations; revise the manuscript; then complete Turnitin, UTUGradu, maturity and examination steps. Reserve contingency time for machine downtime, failed builds and industrial review.

Final pre-submission checklist

Before submission, confirm that the manuscript names KTEK0015 as the 30 ECTS thesis and does not call the whole 40 ECTS category a 40 ECTS thesis; keeps KTEK0026 and KTEK0016 separate; addresses TTDK1308 correctly; identifies process, material, machine and evidence level; reports feedstock and post-processing where relevant; separates sensor signals from physical quality evidence; prevents ML leakage; uses appropriate metrics and grouped validation; distinguishes simulation from measurement; reports uncertainty and replication; avoids turning one successful build into production-readiness claims; resolves NDA, privacy and permit issues; verifies AI-assisted material; and follows current Turnitin and UTUGradu instructions.