Doctor of Philosophy, Case Western Reserve University, 2024, EMC - Mechanical Engineering

In robot control, classical stability is formed around a stable point (attractor) or connected stable points (limit cycles). In contrast, connected saddles can be used to describe stable sequences of states. The connection between two saddles in phase space is a heteroclinic channel, and stable heteroclinic channels (SHCs) can be combined to form cycles and networks – stable heteroclinic networks (SHNs). While the stability and subperiod at each saddle have been mathematically predicted, the potential of SHCs as robot controllers has not been fully realized. To move from modelling to control, tools are needed to more precisely design and manipulate these systems. First, this manuscript expands the SHC-framework with a task space transformation inspired by a popular robot control framework – dynamic movement primitives (DMPs). Stable heteroclinic channel-based movement primitives (SMPs) have an intuitive visualization feature that allows users to easily initialize the controller using only the robot's desired trajectory in its task space. After applying SMPs to a simple robotic system, we characterize the SHC system variables in the larger SMP system, and use the SMP variable nu – the saddle value – for local, real-time controller tuning without compromising the overall stability of the system. Finally, we explore more complex, branching connected-saddle topologies as stable heteroclinic networks. SHCs and SHNs are stochastic systems where noisy external input, such as sensory input, can be used as the stochastic component of the system. For robots, we can use SHNs as a decision-making model where the external input directly drives which decision is made. Overall, this manuscript seeks to parametrize the saddle network frameworks SHCs and SHNs for user-friendly, robust, and versatile robot control. Networks of saddles exist as models for neural activity, neuromechanical models, and robot control, and they can provide further utility in the study and application of (open full item for complete abstract)

Committee: Kathryn Daltorio (Advisor); Roger Quinn (Committee Member); Hillel Chiel (Committee Member); Murat Cenk Cavusoglu (Committee Member) Subjects: Mechanical Engineering; Robotics

Master of Science, The Ohio State University, 0, Industrial and Systems Engineering

Introduction. Grocery store employees who restock shelves are exposed to risk factors for shoulder- and back-related musculoskeletal disorders, due to the nature of the shelf-stocking task, which requires repetitive reaching to shelves that range in height from inches from the floor to overhead while handling products that range widely in weights and shapes. This study investigated the potential for an ergonomic intervention, specifically a prototype height adjustable stocking cart to reduce the physical demands experienced when stocking shelves. This study compared muscle activity, kinematics, and subjective preferences when participants used the prototype cart versus a traditional, manual stocking method. Methods. Twelve subjects, 9 males and 3 females, participated in this study. A traditional stocking method was compared to the prototype cart method for two types of dry grocery products as they were moved to three different destination shelf heights. Normalized 50th and 90th percentile electromyography (EMG) data and maximum kinematic displacements were collected and analyzed. In addition, a questionnaire was used to assess usability. Results. EMG data, kinematic data, and subjective feedback favored the cart prototype over the traditional method. In general, where there were statistically significant effects of method of stocking (prototype cart v. traditional method), muscle activity was lower for the cart condition when transferring products to a high and a low shelf, and there was no effect of stocking method on muscle activity for the middle shelf; this was the finding for the left trapezius, left and right anterior deltoid, and right erector spinae muscles. The cart primarily benefitted the anterior deltoid and trapezius descendens through reduced shoulder flexion, but some conditions also elicited statistically significant differences in the erector spinae and latissimus dorsi. Spine flexion was reduced in the cart condition as a main effect, but sp (open full item for complete abstract)

Committee: Steven Lavender (Committee Member); Carolyn Sommerich (Advisor) Subjects: Industrial Engineering

Master of Science (MS), Ohio University, 2016, Industrial and Systems Engineering (Engineering and Technology)

A significant issue within hospitals is the occurrence and frequency of missing doses. A missing dose is when a medication is not available to be administered to the patient at the required time. Decreasing missing doses can improve the patient's quality of care and decrease cost of care, labor costs, and medication expenses. The objective of this research was to identify the cause(s) of missing doses and provide recommendations on how to reduce the frequency of missing doses. This project analyzed a year of quantitative data to identify conditions that are correlated with missing doses. These conditions were identified so that recommendations could be provided to reduce re-dispenses and interruptions within the medication distribution process. The analysis examined re-dispenses in terms of five different characteristics: time, nursing unit, medication, medication form, and the dispensing pharmacy.

Committee: Dale Masel (Committee Chair); Diana Schwerha (Committee Member); Tao Yuan (Committee Member); Douglas Bolon (Committee Member) Subjects: Engineering; Health Care; Industrial Engineering

MS, University of Cincinnati, 2014, Engineering and Applied Science: Electrical Engineering

There are many feature selection algorithms and many classification methods available to choose from in order to successfully and accurately learn a data set. This work focuses on the merits of dimensionality reduction and the comparative analysis of select techniques. Relief-f, Classification and Regression Trees (CART), and Analysis of Variance (ANOVA) are used to select subsets of features within four real world and one artificial data set. The three are then combined to produce and ensemble feature subset. These results are used as input for feed forward artificial neural networks (FFANN), Naive Bayes, support vector machines (SVM), and an ensemble the three. Averaged accuracy percentages are used to analyze the performance of each dimensionality reduction approach. It was found that the ensemble approach to feature selection produced generally more accurate results overall. Average accuracies included 96.5% of correctly identified benign and malignant breast cancer tumors and 89.5% of appropriately labeled splice junctions within DNA sequences, while reducing the data sets from nine features and sixty features respectively to five representative features each.

Committee: Ali Minai Ph.D. (Committee Chair); Raj Bhatnagar Ph.D. (Committee Member); Carla Purdy Ph.D. (Committee Member) Subjects: Artificial Intelligence

PhD, University of Cincinnati, 2006, Engineering : Computer Science and Engineering

The objective of supervised learning is to estimate unknowns based on labeled training samples. For example, one may have aerial spectrographic readings for a large field planted in corn. Based on spectrographic observation, one would like to determine whether the plants in part of the field are weeds or corn. Since the unknown to be estimated is categorical or discrete, the problem is one of classification. If the unknown to be estimated is continuous, the problem is one of regression or numerical estimation. For example, one may have samples of ozone levels from certain points in the atmosphere. Based on those samples, one would like to estimate the ozone level at other points in the atmosphere. Algorithms for supervised learning are useful tools in many areas of agriculture, medicine, and engineering, including estimation of proper levels of nutrients for cows, prediction of malignant cancer, document analysis, and speech recognition. A few general references on supervised learning include [1], [2], [3], and [4]. Two recent reviews of the supervised learning literature are [5] and [6]. In general, univariate learning tree algorithms have been particularly successful in classification problems, but they can suffer from several fundamental difficulties, e.g., "a representational limitation of univariate decision trees: the orthogonal splits to the feature's axis of the sample space that univariate tree rely on" [8] and overfit [17]. In this thesis, we present a classification procedure for supervised classification that consists of a new univariate decision tree algorithm (Margin Algorithm) and two other related algorithms (Hyperplane and Box Algorithms). The full algorithm overcomes all of the usual limitations of univariate decision trees and is called the Paired Planes Classification Procedure. The Paired Planes Classification Procedure is compared to Support Vector Machines, K-Nearest Neighbors, and decision trees. The Hyperplane Algorithm allows direct user in (open full item for complete abstract)

Committee: Dr. Anca Ralescu (Advisor) Subjects:

MS, University of Cincinnati, 2005, Medicine : Biostatistics (Environmental Health)

Three CART algorithms, CRUISE, GUIDE and LOTUS, were selected to show their equivalency to the traditional statistical method of logistic regression. The experimental data from a postural instability during task performance on elevated and/or inclined surfaces study were used in the analyses. The results of sensitivity, specificity and overall prediction correctness demonstrated that all four methods performed about equally well. Additionally, the receiving operating characteristic curve and the area under the curve calculated for logistic regression and LOTUS were compared and no significant differences were found. CART algorithms displayed the advantage of easier interpretation relative to logistic regression.

Committee: Dr. Paul Succop (Advisor) Subjects: Statistics

Master of Science in Civil Engineering, University of Toledo, 2008, Civil Engineering

This experimental project presents a comprehensive study of indoor pollutant behavior in the public transport buses in the city of Toledo running on alternative fuels and an understanding of the contribution of outdoor pollutant concentrations to in-vehicle pollutant levels. The indoor pollutants monitored are particulate matter, carbon dioxide, carbon monoxide, sulphur dioxide, nitric oxide, and nitrogen dioxide. Temperature and relative humidity are also measured inside the vehicle in addition to the in-vehicle pollutants. The various factors affecting indoor air quality are indoor sources of pollutants (people, furniture, etc.), ventilation, outdoor air quality, meteorology, pollutant decay, and vehicular traffic. The diurnal, monthly, and seasonal variations of the pollutants are studied. The pollutant level buildup within a bus compartment is due to a combination of different factors and not a result of variation due to a single variable. As the bus is in motion and factors influencing the indoor pollutant levels keep changing randomly, it is difficult to identify specific monthly and seasonal trends. However, pollutant concentration levels are found to be highly influenced by peak hours in the morning and evening and a discussion is provided on identifying the factors that could have influenced monthly and seasonal variations. Relatively higher pollutant concentrations are observed for majority of the pollutants in winter when there is not much air exchange in the bus compartment. The trend study revealed that the concentrations were mainly influenced by peak hours, ventilation settings, vehicular traffic, passenger ridership, and meteorology. The factors influencing pollutant levels with respect to month and season are identified. The regression tree analysis helped identify the various factors affecting in-vehicle pollutant levels and the relationships between independent variables and indoor pollutant concentrations. The meteorological effect study revealed (open full item for complete abstract)

Committee: Ashok Kumar PhD (Advisor); Andrew Heydinger PhD (Committee Member); Devinder Kaur PhD (Committee Member) Subjects: Civil Engineering; Environmental Engineering; Transportation

Master of Science (MS), Ohio University, 2009, Industrial and Systems Engineering (Engineering and Technology)

Accuracy is a critical factor in predictive modeling. A predictive model such as a decision tree must be accurate to draw conclusions about a prediction. This research aims at analyzing and improving the performance of classification and regression trees(CART), a decision tree algorithm by evaluating the performance of the algorithm on aset of databases extracted from real world problems. Various methods and parameters of the algorithm were used to develop decision trees. The predictive accuracy of the trees developed by all the methods was examined and a best method that develops a tree with better accuracy was identified. However, a new approach was introduced to further improve the efficiency of the CART algorithm by combining the functionality of CART with neural networks. Neural networks contribute by generating new data necessary to improve the accuracy of the decision trees. Finally, the decision trees developed by both the new method and existing CART were compared for accuracy. This research thus provides insight into improved performance of the CART algorithm, comprehending the behavior of the algorithm and determining methods and parameters for better accuracy.

Committee: Gary R. Weckman PhD (Advisor); Diana Schwerha PhD (Committee Member); Ralph Whaley PhD (Committee Member); Andrew Snow PhD (Committee Member) Subjects: Industrial Engineering