Robust Dynamic Data Driven Adaptive Real-Time Location Tracking System for Workflow Improvement
Localization and tracking
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction, presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
EmbargoRelease after 05/03/2026
AbstractTimely and accurate tracking of workers and materials in a production facility is critical to enhancing its productivity and overall efficiency. Workflow analysis is one of the most widely used approaches to identify improvement opportunities for productivity and efficiency. However, the dynamic and complex nature of production systems makes utilizing traditional workflow analysis very challenging due to the following reasons. First, a detailed workflow analysis of a large production facility requires a massive amount of time measurements for every task. Second, direct observations by human observers for an indefinite time are not feasible and inherently flawed due to human error. Third, data collection based on workers’ input involves bias due to the perceived risk of losing the job. Fourth, there is a lack of algorithms enabling automatic workflow analysis for the production processes that are manual labor-intensive. To address these challenges, novel dynamic data driven frameworks based on real-time location tracking system (RTLTS) are proposed in this dissertation. The RTLTS is based on the passive radio frequency identification (RFID) and ultra-wideband (UWB) sensors that localize and track workers and materials at the macro and micro levels, respectively. The proposed frameworks provide solutions to four major problems that need to be resolved to achieve workflow analysis based on RTLTS at both levels. First, to localize an object utilizing ultra-high frequency (UHF) passive RFID tags, a robust localization framework based on a dynamic K-nearest neighbor algorithm (DKNN) is proposed. The proposed framework collects the received signal strength indicator (RSSI) information by different access points (APs) to generate multiple regression models in order to improve the localization of objects. After validating the generated regression models’ adequacy using Sharpio Wilk, Kolmogorov Smirnov, Cramer-von Mises and Anderson-Darling tests for normality and Levene’s test for the constant variance of error terms, a fingerprinting dataset is then generated. This dataset is used to localize each tag utilizing our proposed DKNN algorithm. This framework iteratively updates our regression models within the localization problem as the environment changes, which makes the framework robust and improves the localization and tracking accuracy accordingly. Second, to track moving objects using UHF RFID tags, a DKNN based hidden Markov model (HMM) is proposed, which estimates the locations of target tags and finds the most likely path of movement. To develop the HMM, the emission matrix is calculated based on the location-likelihood provided by the DKNN, and the transition matrix is developed based on the RSSI values. The Viterbi search algorithm is employed to determine the most likely path. Multiple experiments are conducted to test these proposed frameworks’ performance utilizing different environment types (i.e., obstacles and obstacles), a different arrangement of stationary tags, six different moving patterns, and three different speed levels. The proposed frameworks have achieved 0.36m accuracy for detecting stationary objects and dynamic objects, which outperforms currently available algorithms in the literature. Unlike the traditional HMM with a fixed set of states, the states of the proposed HMM based tracking framework change at each timestamp which is estimated by the DKNN. Therefore, the proposed framework is not only dynamic in nature but also robust enough to accommodate any unforeseen changes over time. Third, an optimal number of readers and APs is determined to maximize the accuracy of localization and tracking. While the accuracy of localization and tracking can be improved by increasing the number of readers and APs to a certain extent, it will cost more for the deployment of the system. Besides, too many APs cause interference and, furthermore, decrease localization and tracking accuracy. Therefore, finding the optimum number of APs required and locations to deploy those for the RTLTS in a given facility without sacrificing the accuracy is important. In this dissertation, a generalized approach is proposed to find the optimum number of APs and their deployment location, considering the circular coverage area by APs. To this end, two algorithms are developed to solve the area covering and tag covering problems. The proposed algorithm can be used for any RTLTS network (i.e., UWB based, RFID based) deployment by changing the set of its input parameters. Fourth, an automatic workflow analysis framework is developed based on RTLTS data. RFID based tracking is cost-effective when a large number of objects need to be tracked for macro-level workflow analysis, such as inventory tracking. For a micro-level workflow analysis, where high tracking accuracy is required, UWB sensors are utilized, which can track the movement with approximately 10cm of accuracy. In this dissertation, an autonomous workflow analysis framework is developed utilizing UWB based RTLTS. The proposed framework is applied to identify worker and task associations in the vegetable grafting process based on the location data. It involves binary segmentation of different distance measures calculated from location data to detect the plant tray’s movements and worker association. An iterative overlap time removal algorithm is then implemented to ensure that no worker is assigned to overlapping processes and no task is processed by multiple workers simultaneously. Key workflow matrices such as flow time, processing time, and waiting times are automatically calculated. Validation of results with video data reveals that the proposed workflow analysis can achieve 90.54% of accuracy. Even though the application of the proposed framework is validated using a small-scale vegetable grafting scenario, it is robust enough to handle other production facilities involving manual labor on a larger scale with little to no changes for automatic workflow analysis. Besides, the methodology can be applied to perform automatic workflow analysis with different tracking technology with similar tracking accuracy of UWB. To the best of our knowledge, the proposed robust RTLTS based approach is one of the first efforts to support the real-time monitoring of the production systems that are heavily labor-intensive such as job shops and assembly lines in manufacturing or agricultural processing facilities. This dissertation focused on addressing the real-world challenges encountered in greenhouse production systems such as grafting nurseries as case studies.
Degree ProgramGraduate College
Systems and Industrial Engineering