Argomenti di tesi disponibili
Siamo un gruppo di ricerca del Politecnico di Milano che opera nel campo della digitalizzazione e Industria 4.0 applicati all’asportazione di truciolo ed in particolare alla fresatura. Riportiamo di seguito le offerte di tesi proposte dal gruppo con alcune note esplicative:
- L’inizio della tesi è sempre trattabile per cui se siete interessati ad un argomento mettetevi comunque in contatto
- Mettete sempre in copia nella mail tutti i contatti presenti per l’argomento selezionato
- I requisiti sono indicativi e, soprattutto per quelli base, facilmente recuperabili durante l’inizio della tesi
- All’inizio della tesi verranno tenute delle lezioni specifiche per formare lo studente direttamente sull’argomento scelto e offrire una panoramica di ampio respiro sulla digitalizzazione in generale
AI in manufacturing: Using computer vision to defect identification in machine tool manufactured products
Abstract: Computer Vision is an artificial intelligence technique for image analysis that is now common industrial research topic in many manufacturing fields. In the field of machine tools, it is a nascent technology with enormous promise in terms of results and use. It is predicted that almost 80% of all datasets used for training AI models in any field will be synthetically generated due to how immensely costly and time consuming it is to create datasets from real world images. The manufacturing sector has enormous difficulties in acquiring real world photographs and images and is also incredibly impenetrable to technological changes. This work as others in this field and lab aims to change this introducing state of the art technologies to solve important problems.
The aim of the thesis is to leverage industrial grade software such as Nvidia TAO (Train, Adapt, Optimize) to create synthetic datasets capable of detecting defects on machined components. Visual inspection and evaluation of both manufactured products and tools is incredibly complex and time consuming. Interrupting production is costly, and human error can still mean that defective parts pass quality inspection. Automatic detection can drastically reduce downtimes and human error.
The thesis will be validated using motorsport components designed by the two racing teams of Politecnico di Milano (PRC Dynamis and PoliMi Motorcycle Factory) and manufactured for them by the group. Real and fabricated defects (utilizing Nvidia TAO) will be added to the component/s and it will be determined if the trained algorithm will be able to detect them accurately and repeatedly.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Moderate/decent level programming in MATLAB or preferably Python, in general a willingness and interest in programming, candidates who have a good preparation in C++ are also welcome. Knowledge of Computer Vision technologies is appreciated but not necessary, if absent it will be provided by the group. Knowledge of manufacturing is welcome but if absent can be provided by the group.
Keywords: Computer Vision, AI, Manufacturing, Python, Quality Control
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
AI in manufacturing: Using computer vision for determining the presence of PPE (Personal Protective Equipment)
Abstract: Computer Vision is an artificial intelligence technique for image analysis that is now common industrial research topic in many manufacturing fields. In the field of machine tools, it is a nascent technology with enormous promise in terms of results and use. It is predicted that almost 80% of all datasets used for training AI models in any field will be synthetically generated due to how immensely costly and time consuming it is to create datasets from real world images. The manufacturing sector has enormous difficulties in acquiring real world photographs and images and is also incredibly impenetrable to technological changes. This work as other in this field and lab aims to change this introducing state of the art technologies to solve important problems.
The aim of the thesis is to leverage state of the art technologies such as Contrastive Language–Image Pre-training (CLIP) provided by OpenAI to generate synthetic datasets to train Computer Vision models on them. This provides a monumental improvement in terms of quality and quantity since specific datasets with real world images of industrial subjects are rare and most often private repositories. The objective is to be able to recognize the presence and position of individuals and ascertain if they are wearing PPE material. This guarantees a safer workplace and ensures that common EU safety laws are being abided by.
The thesis will be validated possibly following a small deployment in a company of choice which habitually collaborates with the PoliMill laboratory by setting up a smart monitoring system around one of their machine tools.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Moderate/decent level programming in MATLAB or preferably Python, in general a willingness and interest in programming, candidates who have a good preparation in C++ are also welcome. Knowledge of Computer Vision technologies is appreciated but not necessary, if absent it will be provided by the group. Knowledge of manufacturing is welcome but if absent can be provided by the group.
Keywords: Computer Vision, AI, Manufacturing, Python, Health & Safety
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
Interaction Design: UI for a software application for machine tools using IIoT
Abstract: The manufacturing sector, and particularly machine tools, is incredibly resistant to change and digitalization. In part this is due to the intrinsic caution companies have in innovating fearing that whilst the digital transformation is taking place machine downtimes will hurt their bottom line. Digital transformation via IoT devices and software is slowly taking place however, adoption is low and often interpretability of the software is scarce, indeed even when deployed these software applications sometimes remain little or seldom used. There are many factors determining this but one of the key ones is usability, the UI and UX of these applications is often badly designed and alienates the user from interacting with it and retrieving easily the key information the application is designed to provide.
The aim of the thesis is to design together with leading researchers in the field and consulting with industrial partners a UX for IIoT (Industrial Internet of Things) devices which is clear communicative and detailed. The UI and UX should implement the newest philosophies in terms of hierarchy, proximity, consistency and so on.
The thesis will be validated by measuring Key Performance Indicators (KPIs) interviewing experts in the manufacturing field and especially users of machine tools. The individuals will be amongst employees of partners of the lab and a larger scale roll out/ test group could be organized during one of the manufacturing events (Workshop/Courses) held by the group.
Study Courses:
- Product Service System Design
- Digital and interaction Design
Prerequisites: Good knowledge of Miro and/or Figma
Keywords: UI, UX, Manufacturing, IoT, Figma
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
IIoT in Manufacturing: Analysis and implementation of an Open-Source IoT communication protocol for retrieving data from machine tools
Abstract: The manufacturing sector has recently been interested by a wave of innovation due to the development of Industry 4.0. One of the most promising and important developments has been the introduction of communication protocols which allow the extraction of information from machine tools to perform data analysis which includes predictive maintenance, energy monitoring. Many machine tool manufacturers have implemented their own proprietary protocols of communication however, the IEC standard committee has published its own standard the IEC 62541 also known as OPC-UA. It is an open source and broadly adopted protocol which can be used not only with machine tools to acquire data but for essentially any kind of mechanized system and is becoming a keystone of automation in manufacturing.
The aim of the thesis is to implement the OPC-UA protocol for a case study on machine tool monitoring and perform data analysis on the acquired data to ascertain present and/or future states of the machine tool. It is possible that the thesis be carried out not only within the university but also on the presence of machine tool resellers and manufacturing companies. Data analysis can also be carried out depending on timings and progress of the thesis work.
The thesis will be validated using motorsport components designed by the two racing teams of Politecnico di Milano (PRC Dynamis and PoliMi Motorcycle Factory) and manufactured for them by the group. Data will be acquired during the machining process of these components and will be compared to data previously acquired from the machine using other less industrial methodologies.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Good knowledge of C/C++ and a willingness to program. Knowledge of QT is preferred but not required since it can be taught during the thesis period by the group. Knowledge of manufacturing is welcome but if absent can be provided by the group.
Keywords:IoT, OPC-UA, C++, Machine Tools
Number of positions: 4
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
AI in Manufacturing: Making machine tool data more accessible by leveraging LLMs
Abstract: Manufacturing is an incredibly complex sector with little known equipment and complex machinery. Machine tools are no exception, indeed, their evolution over almost 200 years and especially in the last 50 has made them intricate electro-mechanical systems. All aspects of data related to machine tools is difficult to interpret and acquire, manuals are often in foreign languages and run for 1000s of pages, the data acquired via IoT devices is often so vast to make its interpretation a lengthy and costly affair.
The aim of the thesis is to leverage RAG (Retrieval Augmented Generation) to train offline LLMs such as LLAMA 3.1 and Mistral NeMo on data acquired directly from machine tools to evaluate performance and accuracy of the results. Part of the thesis may be performed consulting industrial machine tool experts to have an expert opinion of the level accuracy and benchmark the offline models against them. The choice of offline models is a key aspect since it allows for total privacy with extremely sensitive knowledge and know-how, local deployment allows functioning without internet access and direct and unfettered access to data derived directly from machine tools themselves. Furthermore, it is more cost effective for small companies such as those of most of the EU.
The thesis will be validated by benchmarking the results provided by the trained solution against experienced personnel from companies which are partners of the laboratory. The objective is to evaluate accuracy of the response compared to the training dataset and to calibrate the model to avoid hallucinations as much as possible.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Good level programming in MATLAB or preferably Python, in general a willingness and interest in programming, candidates who have a good to high preparation in C++ are also welcome. Previous knowledge of LLM technologies is appreciated but not necessary, if absent it will be provided by the group. Knowledge of manufacturing is welcome but if absent can be provided by the group.
Keywords: IoT, OPC-UA, C++, Machine Tools
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
AI in Manufacturing: A novel way of leveraging AI to allow machine to tools to self-diagnose and report failures and alarms
Abstract: Machine tools are extremely complex. The points of failure are many and can be both hardware and software. Manuals are often in a different language, at best English, and are hundreds if not thousands of pages long. Accurately describing an alarm and the machine’s conditions when it was triggered becomes complex and even expert personnel may get it wrong or take time to diagnose the fault causing expensive downtimes and miscommunication between the supplier and the end user of the machine tool. The objective and scope of the thesis is to leverage the use of LLMs (Large Langue Models) to draw inference not only from the machine tools manuals but also from the control of the machine itself allowing a fusion of data.
The aim of the thesis is therefore to devise a working system that, in a limited number of sample use cases which have the objective initially to only provide proof of concept, allows a chosen LLM to retrieve data from both the machine and its documentation to correctly diagnose and report the fault/alarm opening automatically a ticket with a well-known machine tool reseller/manufacturer.
The thesis will be validated using the portal of a reseller/manufacturer of machine tools to open support tickets automatically. The accuracy of the opened ticket will be evaluated, and the response time compared to that of an expert employee. This is achievable due to the laboratories long held contacts and good standing in the manufacturing industry.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Good level programming in MATLAB or preferably Python, in general a willingness and interest in programming, candidates who have a good to high preparation in C++ are also welcome. Previous knowledge of LLM technologies is appreciated but not necessary, if absent it will be provided by the group.
Keywords: LLM, AI, Python, Manufacturing
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
Digital Manufacturing: Leveraging CL.DATA to predict cutting conditions in CAM programmed machine tools
Abstract: Machine tools are incredibly complex systems. CAM software is used to calculate the toolpath the machine must execute to produce complex parts. CAM software produces a toolpath as an ISO standard language called CL.DATA which the postprocessor then translates into machine specific instructions. This means that reading CL.DATA is easy compared to reading machine specific instructions, it does not require an intimate knowledge of the control and of proprietary commands designed by the manufacturer. This may introduce an element of error but the gains in terms of interoperability and compatibility with all CAM software are immense. Interpolation simulators which are based on CL.DATA are also affected by one major drawback: the lack of knowledge of the acceleration profiles of each axis.
The aim of the thesis is therefore to design an interpolator which is capable of reading CL.DATA, plotting the toolpath of the machining operation, incorporate elements of the dynamics of the machine such as acceleration and deceleration curves. The interpolator must be able to predict the complete machining time.
The thesis will be validated by using as benchmark toolpaths of automotive components deriving from the Dynamis PRC and PoliMi Motorcycle Factory teams. The machining time of the components will be compared between the designed interpolator, the time produced by the CAM, time predicted by and industrial simulator such as VERICUT and the actual machining time. Accuracy will also be considered simply to evaluate the toolpath is being parsed correctly by the interpolator.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: No
Keywords: Manufacturing, CAM, Toolpath, Automotive
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
AI in Manufacturing: The use of Ontologies for the construction of Knowledge Graphs to distill and interpret company knowledge
Abstract: Ontologies are, simply put, a way of interpreting and representing all the multifaceted properties of a subject. The subject may be a specific domain of knowledge or an environment such as a company. Ontologies are based on creating relationships between the various entities of said subject, these entities and relationships are the core of an ontology and can be presented visually in what is called a knowledge graph. The capability of accurately representing the knowledge domain or space of a specific application/environment/topic is an incredibly powerful concept. It allows first and foremost to structure knowledge to make it searchable in a way a simple database or other data repository will never be. It is for this reason that companies like Google, IBM, Microsoft and others all use them to great extent in their company pipelines.
Recently a new implementation method has appeared and that is the use of LLMs and Knowledge Graphs. LLMs (like the one behind ChatGPT) are used to interrogate Knowledge Graphs thus allowing the user to have the information contained within the latter easily and accurately presented, it is a faster and more evolved way of previously developed methods for querying Knowledge Graphs.
The aim of the thesis is to collaborate with a small company in developing a detailed ontology of their machine tools, toolholders and tools. The ontology must reflect their working environment and needs. It must be able to capture and embody their reality as is and allow the company to query the derived knowledge graph in order to be able to extract valuable information.
The thesis will be validated by benchmarking the answers provided by the constructed ontology against the lead answers provided by the lead engineers present at the company. The obtained knowledge graph will be used as a data warehouse to be queried by an LLM and tested to verify that it provides correct answers with respect to the knowledge present in the company.
Study Courses:
- Mechanical Engineering
- Automation Engineering
- Computer Science Engineering
Prerequisites: Average level programming in MATLAB or preferably Python, in general a willingness and interest in programming.
Keywords: Ontology, AI, Knowledg Graph, Manufacturing
Number of positions: 2
Supervisor: Massimiliano Annoni (massimiliano.annoni@polimi.it)
Co-Supervisor: Francesco Barna (francesco.barna@polimi.it)
- LEARNING TIME: -- min.
