Predictive maintenance is an automated technique that analyses sensor data from industrial systems to enable downtime planning. Available for this study is unlabelled data from sensors placed in proximity to hydraulic system outlets of a cold isostatic press. There is limited knowledge about degradation processes because of their rarity, but it is still of high importance to minimise them. One approach to overcome this obstacle is by implementing machine learning to recognise deviations from normal behaviour and potentially learn about them. The state-of-the-art machine learning algorithms for situations with little to no knowledge about anomalies in different machines are deep learning variants using unsupervised learning and transfer learning. With the foundation of such research, this study analyses the available data and proposes three deep learning methods. The testing of these algorithms is made by presenting an equal amount of healthy and simulated unhealthy data as input. The output measurement threshold is adjusted to minimise false negatives because of safety reasons. Consequently, the best method (denoising autoencoder) results in 94% accuracy for separating the data and 74% when also identifying the source of error. However, the results should be taken with caution as the simulated faulty data is not fully representative of a real scenario. These algorithms indicate to what extent they are capable of separating deviations from normal data. This thesis provides knowledge about predictive maintenance and lays a foundation for implementing automatic anomaly detection with deep learning in a high-pressure system.

Object manipulation is a complex task for robots. It often implies a compromise between the degrees-of-freedom of hand and its fingers have (dexterity) and its cost and complexity in terms of control. One strategy to increase the dexterity of robotic hands with low dexterity is called extrinsic manipulation and its principle is to exploit additional accelerations on the object caused by the effect of external forces. We propose a force control based method for performing extrinsic in-hand object manipulation, with force-torque feedback. For this purpose, we use a prehensile pushing action, which consists of pushing the object against an external surface, under quasistatic assumptions. By using a control strategy, we also achieve robustness to parameter uncertainty (such as friction) and perturbations, that are not completely captured by mathematical models of the system. The force control strategy is performed in two different ways: the contact force generated by the interaction between the object and the external surface is controlled using an admittance controller, while an additional control of gripping force applied by the gripper on the object is done through a PI controller. A Kalman filter is used for the estimation of the state of the object, based on force-torque measurements of a sensor at the wrist of the robot. We validate our approach by conducting experiments on a PR2 robot, available at the Robotics, Perception, and Learning lab at KTH Royal Institute of Technology. / Att greppa och manipulera objekt är en komplex uppgift för robotar. Det innebär ofta en kompromiss mellan hand och fingrars frihetsgrader (fingerfärdighet) mot reglersystemets kostnad och komplexitet. Extrinsic manipulation är en strategi för att öka fingerfärdigheten hos robothänder, och dess princip är att utnyttja accelerationer på objektet som orsakas av yttre krafter. Vi föreslår en metod baserad på att reglerakraft för hantering av objekt i handen, genom en återkoppling av kraftmomentet. För detta ändamål använder vi en prehensile pushing action, där objektet puttas mot en yta, under kvasistiska antaganden. Genom att använda en reglerstrategi får vi en robusthet mot parametrars osäkerhet (som friktion) och störningar, vilka inte beskrivs av systemets model. Kraftkontrollstrategin utförs på två olika sätt: kraften mellan objektet och den yttre ytan styrs med en admittance controller medan en ytterligare styrning av applicerad gripkraft på objektet görs med en PI-reglerare. Ett Kalman filter används för att estimera objektets tillstånd, baserat på mätningar av kraftmoment via en sensor vid robotens handled. Vi utvärderar vårt tillvägagångssätt genom att utföraexperiment på en PR2-robot vid KTHs Robotics, Perception och Learning Lab.

Autonomous vehicles face various challenges under difficult terrain conditions such as marginally rural or back-country roads, due to the lack of lane information, road signs or traffic signals. In this thesis, we investigate a novel approach of using Deep Neural Networks (DNNs) to classify off-road surfaces into the types of terrains with the aim of supporting autonomous navigation in unstructured environments. For example, off-road surfaces can be classified as asphalt, gravel, grass, mud, snow, etc. Images from the camera mounted on a mining truck were used to perform semantic segmentation and to classify road surface types. Camera images were segmented manually for training into sets of 16 and 9 classes, for all relevant classes and the drivable classes respectively. A small but diverse dataset of 100 images was augmented and compiled along with nearby frames from the video clips to expand this dataset. Neural networks were used to test the performance for the classification under these off-road conditions. Pre-trained AlexNet was compared to the networks without pre-training. Gabor filters, known to distinguish textured surfaces, was further used to improve the results of the neural network. The experiments show that pre-trained networks perform well with small datasets and many classes. A combination of Gabor filters with pre-trained networks can establish a dependable navigation path under difficult terrain conditions. While the results seem positive for images similar to the training image scenes, the networks fail to perform well in other situations. Though the tests imply that larger datasets are required for dependable results, this is a step closer to making the autonomous vehicles drivable under off-road conditions. / Autonoma fordon står inför olika utmaningar under svåra terrängförhållanden som landsbygds- eller skogsvägar på grund av bristen av körfältinformation, vägskyltar och trafikljus. I denna avhandling undersöker vi ett nytt tillvägagångssätt att använda Djupa Neurala Nätverk (DNN) för att klassificera terrängytor utifrån deras körbarhet i syfte att stödja autonom navigering i ostrukturerade miljöer.Till exempel kan terrängytor klassificeras som asfalt, grus, gräs, lera, snö etc. Bilder från kameran monterad på en gruvbil användes för att utföra semantisk segmentering och klassificera vägytor. Bilderna delades manuellt upp i träningsset på 16 samt 9 klasser för alla relevanta klasser respektive körbara klasser. Ett litet men mångsidigt dataset med 100 bilder förstärktes med närliggande bilder från videoklippen för att expandera detta dataset. Neurala nätverk användes för att testa prestandan hos klassificeringen under dessa terrängförhållanden. Det förtränade nätverket AlexNet jämfördes med nätverken utan träning. Gaborfilter, kända för att särskilja texturerade ytor, användes vidare för att förbättra resultaten av det neurala nätverket. Experimenten visar att förtränade nätverk presterar bra med små dataset och många klasser. En kombination av Gaborfilter med förtränade nätverk kan skapa en pålitlig navigationsväg under svåra terrängförhållanden. Även om resultaten verkar positiva för bilder som liknar träningsbildscenen presterar nätverken inte bra i andra situationer. Även om testen tyder på att stora dataset krävs för tillförlitliga resultat, är detta ett steg närmare att göra de autonoma bilarna körbara i svåra terrängförhållanden.

In manufacturing, emerging digital technologies related to industry 4.0 are playing an assisting role for operators, and just as in previous industrial revolutions the paradigm for operators in the industry is changing. This study has two key goals. The first is to look into the impact of the worker's psychosocial impacts under the operator 4.0 typologies during assembly, training, and maintenance operations, and the second is to look into the potential changes in the operator framework as the industry progresses from 4.0 to 5.0. This study proposed a theoretical framework for assessing psychosocial impacts in operator 4.0 typologies. The proposed framework can be utilized by the company managers, researchers, production engineers, and human resource personnel for the psychosocial risk assessment of the operators in assembly, training, and maintenance operations as self-report questionnaires. This study employed a systematic literature review strategy to answer the study objectives. The findings reveal that the nature of work, the social and organizational environment of work, and individual impacts are all key categories, that might impact operators’ psychosocial environments in assembly, training, and maintenance operations under the operator 4.0 typologies.This study focuses on determining the psychosocial consequences of the operator 4.0 typologies and helps the operators to become more aware, and equipment designers should consider operator psychosocial work conditions when designing new augmented equipment for assisting operators in the work environment. Most advanced technologies are unfamiliar to operators, and they have exhibited a reluctance to accept new technology because it significantly changes their working environment. Which necessitates the training and awareness of operators regarding advanced technologies. Operator 4.0 typologies were introduced with a vision to create a socially sustainable environment for operators. However, the identified psychosocial impacts make it favorable and unfavorable to the operators.

Robots are becoming more common; robotics has gone from bleeding-edge technology to an everyday topic that families discuss around thedinner table.The number of robots in the industry is growing, which means thatthe demand and need for robots to understand the environment it isworking in is also growing.The standard method for a robot to gather information about a sceneinvolves moving to different pre-determined poses from which it canview and analyze the scene. However, this approach does not con-sider the topology of the scene that the robot should explore.This thesis aims to create a two-dimensional approach to determinethe next best view ( 2D-NBV) to view and explore the scene, intro-duced in the method section.The 2D-NBV method converts a point cloud of the scene to an ele-vation map. A segmenting network is used to get the positions ofpre-trained objects. The positions are then used to generate a2DGaussian kernel heatmap of the scene. Using the 2D elevation andGaussian map, the NBV pose is then calculated. The NBV pose isthen converted back to a 6D pose that the robot moves to capture anew point cloud and register it to the scene.The 2D-NBV method is compared to a baseline and a state-of-the-artmethod. The baseline method captures four different point cloudsfrom pre-determined positions and registers them together. The state-of-the-art methods find a point of interest and declare a set of viewcandidates on a sphere around the point. Ray casting is used to findthe pose with the highest information gain. This pose is set as theNBV for the robot to move to. The goal of this thesis is that themethod should perform better than the baseline method, describedfurther in the method section.The evaluation metric used in this thesis is how wellthe differentmethods could estimate the bounding boxes of pre-trained items us-ing an off-the-shelf semantic scene segmentation method. Six sceneswith varying difficulty were constructed to test the methods.The results showed that the 2D-NBV method successfully comple-mented the scene with information about its empty cells. The 2D-NBV outperforms the state-of-the-art on occluded scenes. The 2D-NBV performed overall just as well as the baseline. The reason thatthe 2D-NBV did not outperform the baseline is seen as a consequenceof the information loss going from 3D to 2D.

Vehicular automation is a rapidly advancing field within robotics. These autonomous machines have the potential to perform burdensome and dangerous tasks that historically have been executed by humans which has been a long-time goal for the industry. This thesis aims to develop a computer vision system to enable volume estimation and material classification of the material inside the bucket of an autonomous wheel loader. This information is crucial for autonomous wheel loaders to make decisions. The system is intended to be self-calibrating to ensure future adaptability to different bucket sizes. A Convolutional Neural Network (CNN) based edge detecting network referred to as Dense Extreme Inception Network for Edge Detection (DexiNed) is proposed to both remove redundant information and enhance desired information. By combining the depth perception from a stereo camera and the information extracted from the DexiNed a proposed solution to estimate the volume is presented. A Simple Linear Iterative Clustering (SLIC) approach is applied to extract the material to enable classification of the material. The estimated volume is compared to an annotated true baseline for validation of the system. The thesis presents the precision of the volume estimation and showcases the result of material extraction using three different segment sizes with the SLIC. Additionally, the thesis presents issues concerning material classification.

Quality control is a necessity when it comes to automating a biological lab with the help of robotics. Two major quality control objectives are targeted by the research group PharmBio at Uppsala University: (1) barcode recognition and decoding, and (2) determining the position and orientation of microplates relative the gripper at the end of an industrial robot arm. In order to achieve these objectives, a hardware package with a camera and microcomputer has been designed and built, which can be attached next to the gripper. In addition to the hardware solution, a software stack has been developed and implemented which utilizes the camera and microcomputer to capture digital images. These images are enhanced and processed using machine vision software on the microcomputer, after which the final generated data is sent to an external system for further handling. The final system consisting of integrated hardware and software is capable of achieving both goals: barcode recognition and plate pose determination. However, due to changes in the group’s project plan during the current Master’s project, final implementation of the plate pose determination software remains as future work for a later version. / Kvalitetskontroll är en nödvändighet när det gäller automatiseringen av ett biologiskt lab med hjälp av robotik. Forskargruppen PharmBio vid Uppsala universitet har två huvudsakliga mål gällande kvalitetskontroll: (1) igenkänning och avläsning av streckkoder, och (2) fastställandet av position och riktning av en mikrotiterplatta relativt en gripklo på änden av en robotarm. För att uppnå dessa mål har en enhet innehållandes en kamera och enkortsdator designats och byggts, tänkt att fästas intill gripklon. Utöver denna enhet har ett mjukvarusystem som använder kameran och datorn för att ta bilder utvecklats och implementerats. Dessa bilder behandlas med hjälp av machine vision-mjukvara på enkortsdatorn innan framtagen data skickas vidare till ett externt datorsystem för ytterligare hantering. Det slutgiltiga systemet bestående av integrerad hård- och mjukvara är kapabel att uppnå båda projektmål: streckkodsavläsning och avgöra position/riktning hos en mikrotiterplatta. På grund av ändrad planering hos forskargruppen under arbetets gång kommer dock implementation av mjukvaran framtagen för positions- och riktningsigenkänning dröja till en senare version av projektet.

A case study for this project is performed at Autoliv in Thailand the company uses a lean production flow approach. This manufacturing plant operates on the level of Industry 3.0, which automates processes using information technology. The case study scenario in manufacturing plants that runs in industrial automation 3.0 has some problems in material handling under the logistics division. By implementing AGV (Automated Guided Vehicle) with help of simulation, the time reduction factor through route and time optimization can be processed and displayed. This will have a direct impact on increasing the material handling efficiency in the production plant. The simulation used for AGV in plant layout is Discrete Event Simulation (DES), which divides each event according to its time. Siemens’s Tecnomatix Plant Simulation software is used as a software. This software handles complex production systems and provides control methods. In this thesis work, the entire plant is implemented in the simulated environment based on the plant layout as per dimensions. Then the AGV routing is done from source to destination. The stations are made as per the requirements between the source and the destination for loading from/ unloading to AGV. The processing time of these stations is given as inputs and the simulation is run for a shift to get the throughput. The specification of the AGV such as speed and dimensions given in the simulated environment is taken from the case study of AGV. A total of 55 AGV models’ are studied and 40 of them are selected for this particular plant layout, they are selected based on their features along with the throughput of units transferred from source to destination. The parameters of the AGV are based on the case study of AGV models. The unit throughput acquired in the simulated environment by the AGV is 20% to 22.5% more efficient than manual material handling.

This work proposed a new automated hazard identification (AHI) approach for Pluge&Produce systems. After going through related standards and research works, the required inputs for automated hazard identification were determined. Then, software was presented to demonstrate using these inputs to perform AHI. This software can identify the resource and the emergent hazards. A new approach for identifying the emerging hazards was proposed.This approach uses the concepts of skill types and lookup tables to cover the wide variety of possible hazards when resources work together. Then display all identified hazards in a Hazard Identification Table (HIT). The proposed HIT is designed to support the persons working in risk reduction by drastically reducing the time needed for hazard identification and preparing them to proceed to the next steps in risk analyses.

The greatest feature of using a Baxter robot among other industrial robots is the ability to train this robot conveniently. The training of the robot could be done within a few minutes and it does not need so much knowledge of programming. However, this type of training feature is limited in functionality and needs frequent updating of the software and the license from the manufactural company. As the manufacturer of Baxter Robot no longer exists due to a merger, the thesis has twofold aims, (1) Exploring different functional, installation, calibration, troubleshooting and hardware features of the Baxter robot and (2) demonstrate the functionality of Baxter to perform general tasks of holding, lifting and moving of test objects from one desired position to another position using custom-made programs in Python. Owing to this, details about different software and hardware aspects of Baxter robot is presented in this thesis. Additionally, systematic laboratory tutorials are also presented in appendices for students who want to learn and operate the robot from simple to complicated tasks. In order to keep the Baxter operational for students and researchers in future, when there is no more help available from its manufacturer, this thesis endeavour to cover all these aspects. Thus, the thesis presents a brief understanding of how to use the Baxter Robot in a simple and efficient way to perform a basic industrial task. The kinematics part will show the concepts of forward and inverse kinematics and the DH (the Denavit–Hartenberg) parameters that are important to understand the end-effector position according to the world frame that will give the knowledge of those who are interested in the kinematics part of Baxter robot. The work of the thesis will make it easier to understand how to program a Baxter robot by using Python language and using the simplest way to move the arm to certain positions. The ROS principles, kinematics and Python language programs will provide a good platform to understand the usability of Baxter robot. Furthermore, easy to use laboratory tutorials are devised and are presented in the appendices. These laboratory tutorials will improve the understanding of the readers and provide a step-by-step guide of operating Baxter robot according to the principles of Robotics. In addition to all these points above, the thesis shows useful functions that are built in ROS (Robot Operating System) that make it easier to program the robot in an untraditional way which is one of a contribution of this thesis itself. The usual way to program the robots, in general, is to study the robot kinematics and calculate the position of the end-effector or the tool according to some frame or the world coordinate frame. This calculation can be done by the forward kinematics or the inverse kinematics. The set of programming Baxter robot in this thesis is not the complex calculation of the forward or the inverse kinematics. The tf (transform)tool in ROS has made it easier to reach the joint angles and program Baxter robot using Python.