This study addresses the challenge of mechanism design optimization, particularly focusing on the energy efficiency and design space of reciprocating mechanisms. The research question centers on how to effectively utilize Computer-Aided Design (CAD) simulations alongside Bayesian optimization (BO) and a constrained design space to streamline the design optimization process, overcoming the limitations of traditional kinematic and dynamic analysis methods. The objective is to investigate and develop a novel optimization framework that integrates CAD-based simulations with a BO approach. To achieve this, the study employs a methodological approach. At first, the feasibility of a chosen mechanism design is evaluated through a sequence of CAD-motion simulations to quantify the (in)feasibility of this design. When this design appears to be feasible, a CAD-based design evaluation method is started, in which the objective value is extracted by a sequence of CAD-motion simulations. In this paper, we advocate the use of non-parametric Gaussian processes to build a surrogate model of the objective function and the feasible design space constrained by static and dynamic constraints. The main research results demonstrate that the proposed CAD-based Bayesian optimization framework can effectively identify optimal design parameters that minimize the root mean square (RMS) torque while adhering to specified static and dynamic constraints. This optimization approach significantly reduces the complexity associated with analytic methods, making it scalable to more complex mechanisms and implementable by machine builders. In conclusion, the study successfully develops a novel optimization framework that leverages CAD-based simulations and Bayesian optimization to streamline the design process of mechanisms. The results of an emergency ventilator case study with three design parameters show a reduction of the RMS torque with 71% after 255 CAD-based design evaluations. Moreover, the results demonstrate the effectiveness of incorporating constraints into the design optimization process and the usage of Bayesian optimization (BO), providing insights into the obtained optimum. This approach offers probabilistic insights into the expected improvement of the optimum, highlighting that the full optimization potential is utilized in a computationally efficient manner.

Traditionally, calibration of a galvanometric setup is based upon a mathematical model of an underlying physical reality. These models make a considerable number of assumptions and simplifications. Moreover, they tend to be non-generalizable and lead to non-convex optimization problems encompassing many parameters. Alternatively, several data-driven statistical approaches have been proposed, in which any model of the underlying reality is completely bypassed. The often black-box model is trained purely on the data itself. Although some precautions for overfitting need to be kept in mind, it has been shown that this radically different approach can outperform the traditional mathematical models. On the other hand, some assumptions about the underlying physical truth are both reasonable and simple to implement. We propose to keep the best of both worlds and construct a semi data-driven calibration model with a single built-in assumption: rays exiting the galvanometric setup are straight lines. The data-driven approaches do not exploit this obvious fact. In this work we focus on the intrinsic calibration, i.e. finding the relationship between the input parameters that control the galvanometers and the rays exiting the device. We investigate four different models to predict lines and evaluate them in cross-validation, predicting intersection points on a validation plane and aiming the laser at a specific point in 3D space. We show that our approach outperforms a purely data-driven approach.

Accurate checkerboard detection is of vital importance for computer vision applications, and a variety of checkerboard detectors have been developed in the past decades. While some detectors are able to handle partially occluded checkerboards, they fail when a large occlusion completely divides the checkerboard. We propose a new checkerboard detection pipeline for occluded checkerboards that has a robust performance under varying levels of noise, blurring, and distortion, and for a variety of imaging modalities. This pipeline consists of a checkerboard detector and checkerboard enhancement with Gaussian processes (GP). By learning a mapping from local board coordinates to image pixel coordinates via a Gaussian process, we can fill in occluded corners, expand the board beyond the image borders, allocate detected corners that do not fit an initial grid, and remove noise on the detected corner locations. We show that our method can improve the performance of other publicly available state-of-the-art checkerboard detectors, both in terms of accuracy and the number of corners detected. Our code and datasets are made publicly available. The checkerboard detector pipeline is contained within our Python checkerboard detection library, called PyCBD. The pipeline itself is modular and easy to adapt to different use cases.

Data-driven techniques are finding their way into the calibration procedure of galvanometric setups. However, they bypass the underlying physical or mathematical model completely. Recent work has shown that a simple assumption about an underlying truth can improve the predictions: laser beams leaving the device follow a straight line. In this article, we take that approach a step further. Both the inputs (the pairs of rotations of the two mirrors) and outputs (the straight lines) lie on a manifold. We can incorporate this prior knowledge in the model via constraints built in the formulation of a covariance function. We propose two constrained models: one in which a linear constraint on the direction vector is written as a differential equation, and one in which a quadratic constraint is imposed by a reparametrization of the line coordinates. We compare them with the data-driven and unconstrained model-based approaches. We show that enforcing constraints improves the quality of the predictions significantly and, thus, the accuracy of the calibration. We validate our findings against real world data by predicting points on validation planes, calculating line segment distances, considering the training times for the models, and assessing how much a predicted line resembles an actual straight line.