Integrated circuit (IC) layer topology analysis methods are the main research topic of this doctoral thesis. Multiple methods are presented for IC layer feature analysis along with a software system where they are implemented and tested. Each of different IC layers has distinct features therefore it is very difficult to use universal algorithm for their analysis. A specialized software system was developed to test various analysis algorithms. The system and its architecture is a part of this thesis. Main tasks solved during research of this these were: finding or developing of optimal methods suitable for IC layer structure recognition, software system design and implementation, experimental testing of implemented methods accuracy and efficiency. Thesis consists of introduction, four chapters and the chapter of conclusions. In the introduction scientific novelty of the work is described as well as the aims and tasks of the work are formulated and the author’s publications and structure of the thesis are presented. The first chapter is dedicated to literature review. It covers existing IC layer structure analysis systems and algorithms which are used for this task. Generic image processing and analysis algorithms and methods are also covered as they are used as part of methodology developed in this thesis. The second chapter details different types of IC layers and their properties. Image processing and analysis methods suitable for each of these layer types are... [to full text] / Disertacijoje nagrinėjama integrinių grandynų (IG) topologijos elementų atpažinimo sistemos metodai ir algoritmai, jų taikymas bei pačios sistemos architektūra. Integrinių schemų projektavimo ir gamybos pramonėje problema yra automatinis kiekvieno technologinio lusto sluoksnio vaizdinės informacijos apdorojimas ir analizė, tiksliai išskiriant gamybos proceso metu suformuotas struktūras, tam kad šių duomenų pagalba galima būtų atlikti gamybos proceso tikslumo patikrinimą. Šio disertacijos darbo tyrimų objektas yra puslaidininkinių integrinių schemų sluoksniuose suformuoti elementai. Kiekvieno iš skirtingų sluoksnių struktūros pasižymi skirtingomis savybėmis, todėl labai sunku sukurti universalius analizės metodus. Dėl šios priežasties buvo sukurta speciali programinės įrangos (PĮ) sistema. PĮ architektūra yra disertacijos tyrimų objektas. Pagrindiniai disertacijoje sprendžiami uždaviniai: IG elementų struktūrų atpažinimo metodų pritaikymas ir kūrimas, PĮ sistemos projektavimas ir įgyvendinimas, eksperimentinis įdiegtų metodų efektyvumo ir tikslumo tyrimas. Disertaciją sudaro įvadas, keturi skyriai ir rezultatų apibendrinimas. Įvade nagrinėjamas problemos aktualumas, formuluojamas darbo tikslas bei uždaviniai, aprašomas mokslinis darbo naujumas, pristatomi autoriaus pranešimai ir publikacijos, disertacijos struktūra. Pirmasis skyrius skirtas analitinei literatūros apžvalgai. Jame nagrinėjamos žinomos IG topologijos elementų atpažinimo sistemos ir jose naudojami metodai... [toliau žr. visą tekstą]

In large scale productions of metal sheets, it is important to maintain an effective way to continuously inspect the products passing through the production line. The inspection mainly consists of detection of defects and tracking of ID numbers. This thesis investigates the possibilities to create an automatic inspection system by evaluating different machine learning algorithms for defect detection and optical character recognition (OCR) on metal sheet data. Digit recognition and defect detection are solved separately, where the former compares the object detection algorithm Faster R-CNN and the classical machine learning algorithm NCGF, and the latter is based on unsupervised learning using a convolutional autoencoder (CAE). The advantage of the feature extraction method is that it only needs a couple of samples to be able to classify new digits, which is desirable in this case due to the lack of training data. Faster R-CNN, on the other hand, needs much more training data to solve the same problem. NCGF does however fail to classify noisy images and images of metal sheets containing an alloy, while Faster R-CNN seems to be a more promising solution with a final mean average precision of 98.59%. The CAE approach for defect detection showed promising result. The algorithm learned how to only reconstruct images without defects, resulting in reconstruction errors whenever a defect appears. The errors are initially classified using a basic thresholding approach, resulting in a 98.9% accuracy. However, this classifier requires supervised learning, which is why the clustering algorithm Gaussian mixture model (GMM) is investigated as well. The result shows that it should be possible to use GMM, but that it requires a lot of GPU resources to use it in an end-to-end solution with a CAE.

Software size measurement provides a basis for software project management and plays an important role for its activities such as project management estimations, process benchmarking, and quality control. As size can be measured with functional size measurement (FSM) methods in the early phases of the software projects, functionality is one of the most frequently used metric. On the other hand, FSMs are being criticized by being subjective. The main aim of this thesis is increasing the accuracy of the measurements, by decreasing the number of defects concerning FSMs that are measured by COSMIC FSM method. For this purpose, an approach that allows detecting defects of FSMs automatically is developed. During the development of the approach, first of all error classifications are established. To detect defects of COSMIC FSMs automatically, COSMIC FSM Defect Detection Approach (DDA) is proposed. Later, based on the proposed approach, COSMIC FSM DDT (DDT) is developed.

Terahertz radiation or T-rays or THz radiation refers to the region of the electromagnetic spectrum between approximately 100 GHz and 30 THz. This spectral region is often referred to as the THz gap as these frequencies fall between electronic (measurement of field with antennas) and optical (measurement of power with optical detectors) means of generation. THz measurements may yield useful information about the structural and chemical nature of the material inspected. Examples include detection of voids in materials and protein binding in biomolecules. This report provides an overview of THz measurements of defects in fiber composites. We find that it efficiently detects defects such as voids and delamination in glass fiber composites better than ultrasound, which was widely used for defect characterization in glass fiber earlier. Comparison of the existing methods with THz is presented in the report for characterization of defects.

Quality control is an important part of any production line. It can be done manually but is most efficient if automated. Inspecting qualitycan include many different processes but this thesisisfocusedon the visual inspection for cracks and scratches. The best way of doingthis at the time of writing is with the help of Artificial Intelligence (AI), more specifically Deep Learning (DL).However, these need a training datasetbeforehand to train on and for some smaller companies, this mightnotbean option. This study triesto find an alternative visual inspection method,that does notrelyon atrained deep learning modelfor when trainingdata is severely limited. Our method is to use edge detection algorithmsin combination with a template to find any edge that doesn’t belong. These include scratches, cracks, or misaligned stickers. These anomalies arethen highlighted in the original picture to show where the defect is. Since deep learningis stateof the art ofvisual inspection, it is expected to outperform template matching when sufficiently trained.To find where this occurs,the accuracy of template matching iscompared to the accuracy of adeep learning modelat different training levels. The deep learning modelisto be trained onimage augmenteddatasets of size: 6, 12, 24, 48, 84, 126, 180, 210, 315, and 423. Both template matching and the deep learning modelwas tested on the samebalanceddataset of size 216. Half of the dataset was images of scratched units,and the other half was of unscratched units. This gave a baseline of 50% where anything under would be worse thanjust guessing. Template matching achieved an accuracy of 88%, and the deep learning modelaccuracyrose from 51% to 100%as the training setincreased. This makes template matching have better accuracy then AI trained on dataset of 84imagesor smaller. But a deep learning modeltrained on 126 images doesstart to outperform template matching. Template matching did perform well where no data was available and training adeep learning modelis no option. But unlike a deep learning model, template matching would not need retraining to find other kinds of surface defects. Template matching could also be used to find for example, misplaced stickers. Due to the use of a template, any edge that doesnot match isdetected.  The ways to train deep learning modelis highly customizable to the users need. Due to resourceand knowledge restrictions, a deep dive into this subject was not conducted.For template matching, only Canny edge detection was used whenmeasuringaccuracy. Other edge detection methodssuch as, Sobel, and Prewitt was ruledoutearlier in this study.

In traditional industrial manufacturing, due to the limitations of science and technology, manual inspection methods are still used to detect product surface defects. This method is slow and inefficient due to manual limitations and backward technology. The aim of this thesis is to research whether it is possible to automate this using modern computer hardware and image classification of defects using different deep learning methods. The report concludes, based on results from controlled experiments, that it is possible to achieve a dice coefficient of more than 81%.

The goal of the research conducted for this master's thesis is to understand if active thermography is a suitable technique to detect (identify) and measure (approximate depth and or size) defects in additive manufacturing (AM) processes. Although other non-destructive measurement techniques exist, active thermography is an attractive option for AM applications because of the short measurement times that could be implemented between each layer of a print, and because of the relatively inexpensive equipment required. However, pulse thermography is typically applied to detect larger defects (>1 mm) in materials with high thermal conductivity. It was uncertain if active thermography was sensitive enough to detect the small defects (μm) commonly introduced during AM. Defects of this size are common in AM, and their presence significantly impacts the mechanical properties of the final part. For this reason, the detection limits of active thermography in common AM materials were investigated. Numerical models were created to simulate the heat transfer during active thermography in AM structures (polymer and stainless steel) with defects of varying size. The models included non-ideal conditions such as spectral in-depth absorption of the irradiative pulse and free convection from the object's surface. The spectral properties of acrylonitrile butadiene styrene (ABS), polylactic acid (PLA), and polyamide 12 (PA 12) were measured (see chapter 2) and used in the numerical models. The numerical data indicates that active thermography is sensitive enough to detect the existence of defects smaller than 100 μm in AM materials (see chapter 3). Furthermore, it demonstrates that the defect aspect ratio (defect diameter divided by defect depth) for which traditional 1D thermography models may be used to approximate the depth of defects in 3D systems is approximately 6 (see chapter 4). In addition, the depth of defects with lower aspect ratios (~4) may also be approximated with relatively low error (~10% error). Non-ideal systems (those with convection and spectral in-depth absorption) were simulated, and figures are provided which facilitate the approximation of defect depth using simple, ideal thermography models. Active thermography has shown potential as being an efficient technique for detecting and measuring small defects common in AM.

No description available.

In this thesis we conclude that convolutional neural networks, together with phase-measuring deflectometry techniques, can be used to create models which can detect and classify defects on painted surfaces very well, even compared to experienced humans. Further, we show which preprocessing measures enhances the performance of the models. We see that standardisation does increase the classification accuracy of the models. We demonstrate that cleaning the data through relabelling and removing faulty images improves classification accuracy and especially the models' ability to distinguish between different types of defects. We show that oversampling might be a feasible method to improve accuracy through increasing and balancing the data set by augmenting existing observations. Lastly, we find that combining many images with different patterns heavily increases the classification accuracy of the models. Our proposed approach is demonstrated to work well in a real-time factory environment. An automated quality control of the painted surfaces of Volvo Truck cabins could give great benefits in cost and quality. The automated quality control could provide data for a root-cause analysis and a quick and efficient alarm system. This could significantly streamline production and at the same time reduce costs and errors in production. Corrections and optimisation of the processes could be made in earlier stages in time and with higher precision than today. / I den här rapporten visar vi att modeller av typen convolutional neural networks, tillsammans med phase-measuring deflektometri, kan hitta och klassificera defekter på målade ytor med hög precision, även jämfört med erfarna operatörer. Vidare visar vi vilka databehandlingsåtgärder som ökar modellernas prestanda. Vi ser att standardisering ökar modellernas klassificeringsförmåga. Vi visar att städning av data genom ommärkning och borttagning av felaktiga bilder förbättrar klassificeringsförmågan och särskilt modellernas förmåga att särskilja mellan olika typer av defekter. Vi visar att översampling kan vara en metod för att förbättra precisionen genom att öka och balansera datamängden genom att förändra och duplicera befintliga observationer. Slutligen finner vi att kombinera flera bilder med olika mönster ökar modellernas klassificeringsförmåga väsentligt. Vårt föreslagna tillvägagångssätt har visat sig fungera bra i realtid inom en produktionsmiljö. En automatiserad kvalitetskontroll av de målade ytorna på Volvos lastbilshytter kan ge stora fördelar med avseende på kostnad och kvalitet. Den automatiska kvalitetskontrollen kan ge data för en rotorsaksanalys och ett snabbt och effektivt alarmsystem. Detta kan väsentligt effektivisera produktionen och samtidigt minska kostnader och fel i produktionen. Korrigeringar och optimering av processerna kan göras i tidigare skeden och med högre precision än idag.

La méthode d'identification développée dans cette thèse est inspirée des travaux de D. Rémond. On considérera les données d'entrée suivante : la réponse de la structure, qui sera mesurée de manière discrète, et qui dépendra des dimensions de la structure (temps, espace) le modèle de comportement, qui sera exprimé sous forme d'une équation différentielle ou d'une équation aux dérivées partielles, les conditions aux limites ainsi que la source d'excitation seront considérées comme non mesurées, ou inconnues. La procédure d'identification est composée de trois étapes : la projection sur une base polynomiale orthogonale (polynômes de Chebyshev) du signal mesuré, la différentiation du signal mesuré, l'estimation de paramètres, en transformant l'équation de comportement en une équation algébrique. La poutre de Bernoulli a permis d'établir un lien entre l'ordre de troncature de la base polynomiale et le nombre d'ondes contenu dans le signal projeté. Sur un signal bruité, nous avons pu établir une valeur de nombre d'onde et d'ordre de troncature minimum pour assurer une estimation précise du paramètre à identifier. Grâce à l'exemple de la poutre de Timoshenko, nous avons pu réadapter la procédure d'identification à l'estimation de plusieurs paramètres. Trois paramètres dont les valeurs ont des ordres radicalement différents ont été estimés. Cet exemple illustre également la stratégie de régularisation à adopter avec ce type de problèmes. L'estimation de l'amortissement sur une poutre a été réalisée avec succès, que ce soit à l'aide de sa réponse transitoire ou à l'aide du régime établi. Le cas bidimensionnel de la plaque a également été traité. Il a permis d'établir un lien similaire au cas de la poutre de Bernoulli entre le nombre d'onde et l'ordre de troncature. Deux cas d'applications expérimentales ont été traités au cours de cette thèse. Le premier se base sur le modèle de la poutre de Bernoulli, appliqué à la détection de défaut. En effet on applique un procédé d'identification ayant pour hypothèse initiale la continuité de la structure. Dans le cas où celle-ci ne le serait pas on s'attend à observer une valeur aberrante du paramètre reconstruit. Le procédé permet de localiser avec succès le lieu de la discontinuité. Le second cas applicatif vise à reconstruire l'amortissement d'une structure 2D : une plaque libre-libre. On compare les résultats obtenus à l'aide de notre procédé d'identification à ceux obtenus par Ablitzer à l'aide de la méthode RIFF. Les deux méthodes permettent d'obtenir des résultats sensiblement proches. / The purpose of this work is to adapt and improve the continuous time identification method proposed by D. Rémond for continuous structures. D. Rémond clearly separated this identification method into three steps: signal expansion, signal differentiation and parameter estimation. In this study, both expansion and differentiation steps are drastically improved. An original differentiation method is developed and adapted to partial differentiation. The existing identification process is firstly adapted to continuous structure. Then the expansion and differentiation principle are presented. For this identification purpose a novel differentiation model was proposed. The aim of this novel operator was to limit the sensitivity of the method to the tuning parameter (truncation number). The precision enhancement using this novel operator was highlighted through different examples. An interesting property of Chebyshev polynomials was also brought to the fore : the use of an exact discrete expansion with the polynomials Gauss points. The Gauss points permit an accurate identification using a restricted number of sensors, limiting de facto the signal acquisition duration. In order to reduce the noise sensitivity of the method, a regularization step was added. This regularization step, named the instrumental variable, was inspired from the automation domain. The instrumental variable works as a filter. The identified parameter is recursively filtered through the structure model. The final result is the optimal parameter estimation for a given model. Different numerical applications are depicted. A focus is made on different practical particularities, such as the use of the steady-state response, the identification of multiple parameters, etc. The first experimental application is a crack detection on a beam. The second experimental application is the identification of damping on a plate.