Ciencias Exactas y Ciencias de la Salud
Permanent URI for this collectionhttps://hdl.handle.net/11285/551039
Pertenecen a esta colección Tesis y Trabajos de grado de las Maestrías correspondientes a las Escuelas de Ingeniería y Ciencias así como a Medicina y Ciencias de la Salud.
Browse
Search Results
- Data-driven control of a five-bar parallel robot with compliant joints(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2023-09-01) Ramírez Martínez, Angel; Chong Quero, Jesús Enrique; emimmayorquin; Cruz Villar, Carlos Alberto; Escuela de Ingeniría y Ciencias; Cervantes Culebro, HéctorThis thesis presents a data-driven approach control to minimize trajectory tracking error of a five-bar robot with compliant joints. Adding compliance in the joints introduces modeling and control challenges due to the flexibility. Traditional model-based methods rely on accurate analytical models of the dynamics, which are difficult to obtain for compliant systems. This motivates a data-driven technique that learns online and offline model directly from experiments using vision-based measurements on the physical robot. The core methodology is divided in online training and offline training. The data-driven controller with online training bases its operation on letting the dynamics of the system run using a controller/compensator, in this case it is used a PID. The online training is based on a Neural Network whose input is the Cartesian position of the end-effector, obtained by the vision-based motion capture system. The Neural Network output is the control signal therefore approximates the inverse dynamic model. With this, it is possible to enhanced the control law of the robot to do tracking error minimization. The offline approach involves collecting time-series data capturing the robot's end-effector Cartesian position while moving in its available workspace. The Cartesian position is also obtained by the vision-based motion capture system. These data, which encapsulate the impact of the compliant joints, are used to train a Neural Network to represent the forward dynamics model. The network maps current state and control law inputs to predictions of the next state. Once trained, this Neural Network model is used by an implementation of Model Predictive Control framework to optimize control laws of the two motors to minimize tracking error of a desired end-effector trajectory. At each control step, a finite horizon optimal control problem is solved to find the control signals that minimizes tracking error over a future window. The Neural Network dynamics model is used to predict the outcomes resulting from candidate control actions. Solving this optimization in receding horizon fashion provides feedback correction to reject disturbances. Online training allows the controller to continuously learn from new data, but it relies on the controller used in order to approximate the dynamic system. Nevertheless, online training requires less compute resources and only one thread of execution. On the other hand offline training allows us to train on a fixed dataset all at once, but the implementation requires the existence of a big enough dataset to train the Neural Network, more computation effort due to the optimization problem solution in each sample time, and in this approximation, two or more execution threads to meet the sampling time proposed. Finally, both implementations are compare in order to clearly identify the advantages and disadvantages of each. Throughout this thesis it is presented two methodologies for data-driven modeling and control of compliant robot systems. These approaches could enhance the capability of next-generation compliant and flexible robots designed for safe human interaction and uncertain environments. Overall, the results validate the feasibility and advantages of data-driven methods for controlling compliant robots.
- Detection of suspicious attitudes on video using neuroevolved shallow and deep neural networks models(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2021-11) Flores Munguía, Carlos; Terashima Marín, Hugo; puemcuervo/tolmquevedo; Oliva, Diego; Ortiz Bayliss, Jose Carlos; School of Engineering and Sciences; Campus MonterreyThe analysis of surveillance cameras is a critical task usually limited by the people involved in the video supervision devoted to such a task, their knowledge, and their judgment. Security guards protect other people from different events that can compromise their security, like robbery, extortion, fraud, vehicle theft, and more, converting them to an essential part of this type of protection system. If they are not paying attention, crimes may be overlooked. Nonetheless, different approaches have arisen to automate this task. The methods are mainly based on machine learning and benefit from developing neural networks that extract underlying information from input videos. However, despite how competent those networks have proved to be, developers must face the challenging task of defining the architecture and hyperparameters that allow the network to work adequately and optimize the use of computational resources. Furthermore, selecting the architecture and hyperparameters may significantly impact the neural networks’ performance if it is not carried out adequately. No matter the type of neural network used, shallow, dense, convolutional, 3D convolutional, or recurrent; hyperparameter selection must be performed using empirical knowledge thanks to the expertise of the designer, or even with the help of automated approaches like Random Search or Bayesian Optimization. However, such methods suffer from problems like not covering the solution space well, especially if the space is made up of large dimensions. Alternatively, the requirement to evaluate the models many times to get more information about the evaluation of the objective function, employing a diverse set of hyperparameters. This work proposes a model that generates, through a genetic algorithm, neural networks for behavior classification within videos. The application of genetic algorithms allows the exploration in the hyperparameters solution space in different directions simultaneously. Two types of neural networks are evolved as part of the thesis work: shallow and deep networks, the latter based on dense layers and 3D convolutions. Each sort of network takes distinct input data types: the evolution of people’s pose and videos’ sequences, respectively. Shallow neural networks are generated by NeuroEvolution of Augmented Topologies (NEAT), while CoDeepNEAT generates deep networks. NEAT uses a direct encoding, meaning that each node and connection in the network is directly represented in the chromosome. In contrast, CoDeepNEAT uses indirect encoding, making use of cooperative coevolution of blueprints and modules. This work trains networks and tests them using the Kranok-NV dataset, which exhibited better results than their competitors on various standard metrics.
- Design and implementation of a quantum multilayer neural network framewori(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2020-12) Gamboa Vázquez, Ariel Arturo Goubiah; HERNANDEZ GRESS, NEIL; 21847; Hernández Gress, Neil; puemcuervo; Aspuru Guzik, Alan; González Mendoza, Miguel; School of Engineering and Sciences; Campus MonterreyArtificial Neurons are biologically inspired algorithms that form the building blocks for Artificial Neural Networks (ANNs) and Multilayer Neural Networks (MNN), which have been recently studied and implemented to solve important ptoblems. Advances in Learning theory and the availability of powerful computational systems has resulted in the creation of many real-world applications. Practically every industry has already adopted Multilayer Learning powered technologies in some part of their processes, as state of the art MNNpowered algorithms can outperform other algorithms and even human accuracy for a wide number of tasks. However, their performance relies heavily on the budget of data available as well as its format, as the most popular applications require a copious amount of training examples. Another limitation to build large scale MNN applications is the vast computational resources needed to build these systems. MNN based algorithms usage is widespread and also getting more complex, this phenomenon creates an ever-growing demand for computational power, which may no longer be satisfied at some point in the new future, thanks to the deceleration in state of the art monolithic processors’ performance. Quantum information theory, is a field that has had success in the last couple of decades, thanks to the creation of algorithms that are in theory able to outperform classical computers. The ability of quantum computers of working with inherently different physical systems than the ones used by classical computers, opens an exciting opportunity for scientists and companies to explore the performance of quantum computers for machine learning tasks, being multilayer learning a focus point, thanks to its importance in classical computing. Although a considerable amount of resources have been allocated to the development of MNN powered algorithms in quantum computers, there are still challenges left to overcome before Quantum Multilayer Neural Networks come to be a technology that can compete with state of the art MNN powered algorithms. This research explores the properties of multilayer neural network algorithms running on quantum computers. The first contribution of the research work reported in this document is the analysis and implementation of a perceptron algorithm running on a quantum computer. The second contribution is the proposal, implementation and analysis of two different information encoding methods for quantum computers. The final, and most important contribution of this work, is the development of a framework that allows training multilayer neural networks for Supervised Learning.
- Detection of Violent Behavior in Open Environments Using Pose Estimation and Neural Networks(Instituto Tecnológico y de Estudios Superiores de Monterrey) Chong Loo, Kevin Brian Kwan; TERASHIMA MARIN, HUGO; 65879; Terashima Marín, Hugo; tolmquevedo, emipsanchez; Conant Pablos, Santiago Enrique; Escuela de Ingeniería y Ciencia; Campus MonterreyPeople’s safety and security have always been an issue to attend. With the coming of techno- logical advances, part of it has been used to improve safeguards, though other aspects, without precautions, have made people even more vulnerable. People can get their sensitive data stolen or become victims of transaction fraud. These may be crimes done without physical interac- tion, but felonies with physical violence still exist. Some solutions for pedestrian safety are guards, police cars patrolling, sensors and security cameras. Nonetheless, these methods only react when the crime is happening or, even more critical, when it has already occurred, and the damage has been done. Therefore, numerous methods have been implemented using Arti- ficial Intelligence in order to solve this problem. Many approaches to detect violent behavior and action recognition rely on 3D convolutional neural networks (3D CNNs), spatial tempo- ral models, long short term memory networks, pose estimation among other implementations. However, in the current state of the art, how these approaches are used do not work perfectly and are not adapted to an uncontrolled environment. Therefore, a significant contribution from this work was the development of a new solu- tion model that is able to detect violent behavior. This approach focuses on using pedestrian detection, tracking, pose estimation and neural networks to predict pedestrian behavior in video frames. This method uses a time window frame to extract joint angles, given by the pose estimation algorithm, as features for classifying behavior. At the moment of developing this thesis project, there were not many databases with violent behavior videos. The ones that existed were low quality; cluttered were pedestrians cannot be seen clearly, and with unfixed camera angles. Consequently, another important contribution of this work was creating a new database, Kranok-NV, with a total of 3,683 normal and violent videos. This database was used to train and test the solution model. For the evaluation, a protocol was designed using 10-fold cross- validation. With the implemented solution model, accuracy of more than 98% was achieved on the Kranok-NV database. This approach surpassed the performance of state of the art methods for violence detection and action recognition in the developed database. Though this new solution model is able to detect violent and normal behavior, it can be easily extended to classify more types of behaviors. Further work requires to test this approach in emerging databases of videos and optimize specific areas of the solution model. Additionally, the contributions of this work can aid in the development of new approaches.