WO2024081676A2 - Automated workflow analysis using location tracking - Google Patents

Abstract

A framework for workflow analysis is developed based on a real-time indoor location tracking system that can utilize Ultra-Wide Band technology to identify worker and task associations in a labor-intensive production process, based on location data. The framework includes binary segmentation of different distance measures calculated from location data to detect an object's movements and worker association in production process. An iterative overlap time removal algorithm can be implemented to ensure that no worker is assigned to an overlapping processes, and no task is processed by multiple workers simultaneously. Key workflow metrices such as flow time, processing time, and waiting times can be automatically calculated; and associated reporting provides valuable insights into the workflow to identify improvement opportunities.

Description

Attorney’s Docket No.: 085067-773789 UA20-202 AUTOMATED WORKFLOW ANALYSIS USING LOCATION TRACKING GOVERNMENT SUPPORT This invention was made with support from the U.S. Department of Agriculture (USDA) National Institute of Food and Agriculture under project number 2016-51181-25404. The government has certain rights in the invention. CROSS REFERENCE TO RELATED APPLICATIONS This is a PCT application that claims benefit to U.S. Provisional Application Serial No.63/414,780, filed on October 10, 2022, which is herein incorporated by reference in its entirety. FIELD The present disclosure generally relates to tracking systems and computerized workflow analysis; and in particular to an inventive framework for workflow analysis using real- time location tracking system data. BACKGROUND Workflow refers to the sequence of tasks performed by workers or machines either independently or in collaboration to achieve the final output. Information on workflow is key to improving efficiency of any production facility. However, a detailed analysis of workflow is a time and task intensive process due to the complexity involved in the data collection phase. Traditionally, data collected for workflow analyses are taken by a human observer who captures various streams of data in the environment of interest. However, this process is not only expensive and unable to operate indefinitely but also is inherently flawed with human errors induced during the data collection process. It is with these observations in mind, among others, that various aspects of the present disclosure were conceived and developed. 1 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 SUMMARY The present disclosure provides a number of examples that describe a framework for real-time indoor location tracking that can utilize Ultra-Wide Band technology to identify worker and task associations in a labor-intensive production process, based on location data. In the context of the disclosed methods, devices, techniques, apparatus, systems, and so on, the terms “operable to,” “configured to,” and “capable of” used herein are interchangeable. In a first set of illustrative examples, the framework described herein is embodied by a system for real-time tracking system with automated workflow analysis based on location tracking, comprising a set of tracking devices including a plurality of tags configured for engagement with objects and workers of a manufacturing process, wherein the set of tracking devices generate signals indicative as to a location of the workers and objects relative to predetermined locations associated with tasks of the manufacturing process; and a processor having access to the signals and in operable communication with a memory. The memory stores instructions the processor executes to: preprocess the signals to generate location data that tracks the objects through the manufacturing process over a time period in a predetermined format, detect changepoints in the location data during the time period using at least a distance between the workers and the objects, and compute one or more workflow metrics based on differences between two or more of the changepoints to illuminate possible adjustments for improving the manufacturing process. In a second set of illustrative examples, the framework described herein is embodied by a method executable by a processor for accessing location data from a set of more tracking devices (of a real-time locating system (RTLS)), preprocessing the location data to identify locations of a tag associated with an object or worker over a time series, detecting changepoints via movement of an object relative to a worker and/or predefined origin, and conducting workflow analysis to compute one or more metrics based on differences between two or more of the changepoints to illuminate possible adjustments for improving the manufacturing process. The foregoing examples broadly outline various aspects, features, and technical advantages of examples according to the disclosure in order that the detailed description that follows may be better understood. It is further appreciated that the above operations described in the context of the illustrative example method, device, and computer-readable medium are not required and that one or more operations may be excluded and/or other additional operations 2 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 discussed herein may be included. Additional features and advantages will be described hereinafter. The conception and specific examples illustrated and described herein may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Such equivalent constructions do not depart from the spirit and scope of the appended claims. 3 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 BRIEF DESCRIPTION OF THE DRAWINGS FIG.1A is a simplified block diagram of an example system for supporting a framework for real-time indoor location tracking that can utilize Ultra-Wide Band technology to identify worker and task associations in a labor-intensive production process, based on location data. FIG.1B is an illustration of material flow and tasks for an example implementation of the system of FIG.1A or framework described herein in a vegetable grafting process. FIG.2 is a simplified illustration of a computer-implemented workflow analysis framework as referenced and described herein. FIG.3 is a series of images of example hardware associated with the framework of FIG.2. FIG.4 is an illustration of an example setup for small and large networks of a tracking system showing different example components and communication media. FIG.5 is a flow diagram and example data associated with data pre-processing steps for the framework described herein. FIGS.6A-6C are example reports showing multiple changepoint detection using binary segmentation including (6A) sample output, (6B) changepoint detection plot, and (6C) penalty value vs. the number of changepoints. FIGS.7A-7C are illustrations of an example worker overlap time removal process for a tray including (7A) output of movement detection unit, (7B) outcome if ^^₁₁ > ^^₂₁, and (7C) outcome if ^^₁₁ < ^^₂₁. FIGS.8A-8B is an illustration of a tray overlap time removal process for tray-1 and tray-2; (8A) output after worker overlap time removal from both trays, and (8B) outcome after tray overlap time removal if ^^₁₁ > ^^₁₂, ^^₂₁ < ^^₂₂, ^^ ^^ ^^ ^^₃₁ > ^^₃₂. FIGS.9A-9D are illustrations and images associated with an example real-time indoor location tracking system (RTILTS) workflow study in a vegetable growing facility; (9A) layout of the vegetable grafting facility showing table arrangement for grafting, location of different departments and ultra-wideband (UWB) anchors, (9B) a table arrangement in the vegetable grafting facility, and (9C) a UWB anchor mounted on the wall, and (9D) a gateway mounted on the wall for data collection. 4 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 FIGS.10A-10C are illustrations of different example tasks of grafting operations; (10A) cutting scions/rootstocks, (10B) clipping rootstocks, and (10C) joining scions and rootstocks. FIGS.11A-11E are illustrations of an example RTLIS web interface; (11A) showing real-time location tracking including a list of UWB gateways, anchors (triangle-shaped components), and tags (shown with circles) in four sections, (11B) a magnified view of section 1, (11C) a magnified view of section 2, (11D) a magnified view of section 3, and (11E) a magnified view of section 4. FIGS.12A-12B are graphical images illustrating location plots for tray tags including (12A) a scatter plot of all locations for all trays, and (12B) a location plot for one tray after being processed by the inventive framework described herein. FIG.13 is a bar plot showing different workflow times, total processing time, waiting time, and flow time for each tray. FIGS.14A-14B are Gantt charts of processing times including (14A) tray vs. workers’ timeline, and (14B) worker vs. the trays’ timeline. FIG.15 is a bar plot showing different workflow times, total processing time for each task in sequence, and different workers that perform each task. FIGS.16A-16E are graphs and plots associated with outlier removal and Tukey’s honestly significant difference (HSD) test. FIG.17 is an error plot from different trays and different tasks. Corresponding reference characters indicate corresponding elements among the view of the drawings. The headings used in the figures do not limit the scope of the claims. 5 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 DETAILED DESCRIPTION Aspects of the present disclosure relate to computer-implemented systems and associated methods for a novel workflow analysis framework configured to analyze the movement of objects (i.e., humans, materials) based on RTILTS utilizing e.g., UWB technology. More specifically, the framework includes a real-time indoor location tracking system that can utilize Ultra-Wide Band technology to identify worker and task associations in a labor-intensive production process, based on location data. The framework includes binary segmentation of different distance measures calculated from location data to detect an object’s movements and worker association in production process. An iterative overlap time removal algorithm can be implemented to ensure that no worker is assigned to an overlapping processes, and no task is processed by multiple workers simultaneously. Key workflow metrices such as flow time, processing time, and waiting times can be automatically calculated; and associated reporting provides valuable insights into the workflow to identify improvement opportunities. In some examples, UWB based RTILTS his implemented by the framework due to low power consumption, less susceptibility to other radio signals, and the ability to differentiate between correct and reflected or refracted signals. Therefore, UWB can achieve high indoor location accuracy with the precise time of arrival measurement, making it an appropriate RTILTS for workflow analysis, where high location tracking accuracy is critical. The continuous stream of UWB based RTILTS time series can be leveraged to analyze, e.g., data of plant trays and workers’ locations. The location data can then transformed through multiple steps to measure the workflow related metrices. In one non-limiting example, the proposed framework is deployed to monitor the flow of material and personnel in a vegetable grafting facility by monitoring the movement of plant trays and workers. Vegetable grafting is a horticultural plant production process of improving crop yields by acquiring resistance against diseases and pests. In vegetable grafting, a plant that has the greater yield and/or more desirable fruit properties is used for the upper or harvestable plant portion (i.e., scion), while a plant that is more resistant to pests and diseases is used for the lower part that roots into the ground (i.e., rootstock). The vegetable grafting process consists of four major tasks: 1) cutting the scion seedling, 2) cutting the rootstock seedling, 3) placing a clip of the right size over the cut end of the rootstock seedling, and 4) firmly joining the scion and rootstock together. Each of these tasks requires a high level of worker expertise and 6 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 sometimes specific equipment to enhance both speed and success rate. Although plant health and compatibility of scion and rootstock are key factors for successful vegetable grafting, the role of human capital is also significant. Besides, the time spent during each task and time to finish all these four tasks also affect vegetable grafting success. Therefore, understanding workflow is critical to improving the productivity of a vegetable grafting facility. The proposed workflow analysis framework calculates different workflow metrices and can automatically generate Gantt charts to visualize and detect any bottlenecks arising from imbalanced worker efficiency. This framework is autonomous and can be used for continuous workflow monitoring and analyses to improve the vegetable grafting processes. It is believed the present RTILTS-based workflow analysis framework is one of the first efforts to support the real-time monitoring of the production systems that are heavily labor-intensive such as vegetable grafting. This framework can also be applied to job shops and assembly lines in manufacturing for workflow analysis. Introduction As indicated herein, information about workflow is key to improving efficiency of any production or manufacturing facility. A real-time indoor location tracking system (RTILTS) can be used to automate workflow monitoring generally. An RTILTS enables the user to obtain real-time localization information of static or dynamic objects from inside or outside. Even though an RTILTS is designed to retrieve each objects’ locations precisely, there are some additional benefits realized by using this location information appropriately. For instance, location data from RTILTS sometimes contains valuable contextual information such as production sequence, layout information, and material handling time. At present, there are many RTILTS technologies available, such as wireless (i.e., Zigbee, RFID, UWB, WiFi, and Bluetooth), infrared, ultrasonic, and computer vision. However, determining the appropriate technology for productivity analysis depends on the type of application, localization accuracy, frequency of update, and work environment. Among different RTILTS technologies, UWB provides better localization and tracking accuracy than others (i.e., as low as 1 cm). Some applications of UWB based RTILTS in different areas besides location tracking are automating workflow in construction, improving healthcare by tracking patients, and calculating operation speeds and times in manufacturing. Even though there are some direct benefits of RTILTS on improving workflow, the application area remains limited. Most of the RTILTS literature is focused on improving localization accuracy. However, improvement of 7 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 localization accuracy can only be realized if it can be related directly to productivity improvement and returns that can offset the cost of deployment. Proposed Framework and Methodologies Referring to FIG.1A, a general system 100 is illustrated that is configured to support the framework described herein for real-time indoor location tracking (using, e.g., Ultra-Wide Band (UWB) technology) to identify worker and task associations in a labor-intensive production process, based on location data. In general, the system 100 includes at least one of a processor 102 or processing element and at least one of a memory 103 or storage device storing instructions 104 accessible by the processor 102 to perform various functions and operations described herein. The system 100 can further include a network interface 106 or multiple network interfaces, and a bus or wireless medium (not shown) for interconnecting the aforementioned components. The network interface 106 includes the mechanical, electrical, and signaling circuitry for communicating data over links (e.g., wires or wireless links) within a network (e.g., the Internet). The network interface 106 may be configured to transmit and/or receive data using a variety of different communication protocols, as will be understood by those skilled in the art. In addition, the system 100 includes a real-time indoor location tracking system, or RTILTS 110 having a set of tracking devices 112 that generates signals 114 accessible by the processor 102 about movement and location of objects and workers associated with a production facility 120. As indicated, the production facility 120 can define a number of predetermined stations 122. Stations 122 can include, by example, predetermined locations along the production facility 120 that correlate to specific tasks of a given production process. Tracking devices 112 include, by non-limiting examples, one or more tags, anchors, gateways, or listeners. Tracking devices 112 identify and track the location of objects or people in the production facility 120 (in real time). Tags can be wireless devices attached to objects or worn by people, and anchors can provide fixed reference points that receive wireless signals from tags to determine tag location over a time prior or time series. The instructions 104 can be implemented as code and/or machine-executable instructions executable by the processor 102 that may represent one or more of a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, an object, a software package, a class, or any combination of instructions, data structures, or program statements, and 8 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 the like. In other words, the instructions 104 or any operations performed by the processor 102 described herein may be implemented by hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware or microcode, the program code or code segments to perform the necessary tasks (e.g., a computer-program product) may be stored in a computer-readable or machine-readable medium (e.g., the memory 103), and the processor 102 performs the tasks defined by the code. In general, the processor 102, as configured by the instructions 104 or otherwise, accesses the signals 114 generated by the tracking devices 112 to generate location data and further information to identify worker and task associations in a labor-intensive production process, based on location data, as further described herein. Referring to FIG.1B, material flow is key to developing a successful workflow analysis framework via RTILTS. A non-limiting example is provided investigating a vegetable grafting facility that produces high-quality grafted vegetable seedlings. In the production process, plants go through multiple operational steps, as shown by the example workflow 150 of FIG.1B. A detailed description of each step can be found in Masoud et al. While much of the inventive framework is described with reference to the workflow 150, the workflow 150 is non- limiting and the framework (for workflow analysis using real-time location tracking system data) described herein can be applied to any number or type of production or manufacturing process. A general workflow for a production process includes various tasks, often associated with predetermined locations at different stations of a production facility. As shown by the flow of material in a plant production process in FIG.1B, the seedlings for both scion and rootstock are ready to be grafted (152) after germination and growing steps. The production process of grafting 152 includes various tasks, designated tasks 154. At the grafting (152) step, scion and rootstock plants grown in individual trays (25 an x 51 cm) are first cut at the same angles at specific position of the stem (scion cutting task 152A and rootstock cutting task 152B). Then, clips are placed on the cut stem of the rootstock seedlings (clipping task 152C). Since all seedlings don’t grow uniformly during the pre-grafting growing stage, the worker assigned to put clips on the rootstock needs to find the clip’s appropriate size based on visual observation. After that, the end cuts of both scion and rootstock seedlings are joined together (joining task 152D). Same as the clipper, the joiner needs to find the proper match between cut scion and cut rootstock seedlings for joining. All of these four tasks 154 are not only very labor-intensive but 9 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 also require visual expertise. Next, the grafted plants are placed in the humidity-controlled room for healing so that both rootstock and scion can grow together as one plant. After acclimatizing, trays with grafted plants are taken out of the humidity-controlled room and then they are moved to a greenhouse for 1-2 more weeks before being packaged and shipped. In illustrating an example implementation of the system 100, the focus is narrowed to the example vegetable grafting process 152. Typically, one worker is responsible for doing one task in the vegetable grafting process. In this example application, one goal of the workflow analysis (based on continuous stream of RTILTS time series data from the RTILTS 110) is to identify how much time the grafted seedlings spend during each step or task, the waiting time between those steps, and which worker is assigned to the task. Referring to FIG.2, a framework 200 supported by the system 100 is configured for such workflow analysis. In one general example, the framework 200 includes four units 202 and a database 204. These units 202 include: 1) a UWB-based RTILTS UWB system unit, designated UWB system unit (202D) (e.g., a specific UWB-version of the RTILTS 110 in FIG.1). The units 202 of the framework 200 further include a 2) a data pre- processing unit (202A), 3) a movement detection unit (202B), and 4) workflow analysis unit (202C). The units 202A-202C can define code and/or machine-executable instructions executable by the processor 102 to perform operations described herein in view of location information received from the UWB system unit 202D (e.g., signals 114). The UWB system unit 202D can collect the location data from different tags (implemented along objects or workers within the production facility 120 of FIG.1), which can be saved into the database 204. The processor 102 then executes the data pre-processing unit 202A and pre-processes this location information to separate different tray information, replace missing values, and calculate distance from the origin of the RTILTS network. After pre- processing the data, the data can be segmented by the processor 102 based on location and distance from a predefined origin. The processor 102 then executes the workflow analysis unit 202C and uses this segmented data to calculate different workflow metrices, such as processing time at each vegetable grafting step, waiting time, and flow time. The unit also accommodates automatic generation of the Gantt charts and bar charts from the results to visualize the workflow. A detailed description of each unit and its functions is shown in FIG.2 (and referenced herein). 10 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 UWB System Unit (202D) Typically a UWB based RTILTS system has three to four components depending on the network’s size, which can include: 1) tags, 2) anchors, 3) listeners, and 4) gateways. A radio wave signal is sent back and forth from the moving tag to the stationary anchor, and a measurement of the time of flight is recorded. Time of flight is the measurement of the time taken by the signal to make the trip by two way ranging. The value of the distance between the anchors and tags is obtained simply by dividing the time of flight measured from the anchor by the speed of radio waves (speed of light c = 3 × 10⁸ m/s) (Dabove et al., 2018). Utilizing the distance information from multiple anchors, the location is calculated via the triangulation algorithm. Aspects of the framework 200 can implement the Decawave MDEK1001 module, which consists of the DWM1001 UWB module, as shown FIG.3. Each UWB module can be configured as an anchor, or tag, or listener. For a large RTILTS network, gateways are needed. A gateway can be developed by combining the Raspberry PI and DWM1001 development board via input—output header. FIG.3 shows the gateway produced by us for the UWB based RTILTS or UWB system unit 202D of the framework 200. To set up a RTILTS network of the UWB system unit 202D, at least an anchor needs to be configured as an initiator, which will initiate the network. The number of anchors needed for the RTILTS network depends on the coverage area, non-line of sight condition, and obstacles. Depending on those conditions, an RTILTS network can be small or large. For instance, if objects need to be tracked within a department without any separation by walls, a small network with a few anchors is enough. However, if objects need to be tracked in two different departments separated by walls, a large network setup is needed. For some examples of the proposed framework 200 (e.g., vegetable grafting), a large network was used as there are multiple departments that are separated by walls or thick curtains. FIG.4 shows the typical small and large UWB based RTILTS network setup. Here, information between anchors and tags is exchanged via UWB. For a small network, the location information of different tags is saved via the listener directly connected to the local computer. If the RTILTS network is large, gateways are needed to collect the location information. Gateway is able to send the location information via an internet server to a remote computer within the same network formed by either ethernet or Wi-Fi. The configuration of the UWB modules can be changed over Bluetooth connection via the android app provided by Decawave. 11 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 Data pre-processing unit (202A) The data pre-processing unit configures the processor 102 to perform five different steps to clean, configure, and format signals 114 from the tracking devices 112 or other data accessible from the tracking devices 112 in the desired way so that the subsequent units can easily analyze it. These steps include: 1) reading data, 2) origin distance calculation, 3) data separation, 4) keeping one record and data association, and 5) missing value replacement. The first step of data preprocessing is reading data. FIG.5 shows the outcome of the data preprocessing steps. The location information from the tracking devices 112 of the RTILTS 110 (or UWB unit 202D) is saved as a text file. In this step, the processor 102 parses the text file and creates a data frame with different data columns. Additionally, the processor 102 removes the NaN values and formats different columns depending on the data types (i.e., numeric, date-time, text). FIG.5 shows the output of reading data where each line of the text file is converted to a row of the data frame with the appropriate column names and data types. In the next step, the distance from a predefined origin is calculated. While setting up the RTILTS network, an origin needs to be defined within the coverage area. The location of the tags in cartesian coordinates are calculated based on this origin. The distance is useful for detecting location change. Since RTILTS 110 collects all tag information simultaneously, the data can be separated based on tagID. In the example of FIG.2, the tags are used to track objects in the form of plant trays and workers. Sometimes, after placing the grafted plant tray in the healing department, it is reused to track another plant tray. Therefore, in the data separation step, the data are separated into different data frames depending on the tagID and reusing status. This step utilizes the distance from origin information calculated in the previous step to separate the reused tags. FIG.5 shows the distance vs. time plot for tagID 9cae, which was reused. At some point, the tag was moved quite far away from the origin. In this vegetable grafting operation, this movement refers to the time when the tray was placed in the healing department. After that, the tag was removed for tracking the next tray. For the application of FIG.2, a threshold of 1800 cm was set to separate the data. In this application, the coordinate of the closest corner of the shelves for placing the grafted plants was ( ^^, ^^) = (15 ^^, 10 ^^) with respect to the origin. The distance of this point is 18.02m from the origin. Therefore, if the tag is more than 1800 cm away from the origin, it means the tray was placed in the healing department after finishing the vegetable grafting. Since the proposed 12 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 framework 200 is based on the RTILTS information, this threshold is set solely based on the coordinate measurement of the shelves at the healing department; and the data are separated into two data frames accordingly for each tag. UWB systems can collect data from tracking devices as much as ten times per second. In vegetable grafting, there are no such movements that require tracking at the millisecond level. Therefore, it can be safely assumed that, within one second period, there will be no change in a tag’s location. Storing one record and data association step keeps the last location data recorded at each second. This step eliminates any redundancy and noises to some extent. It reduces the number of data points and makes the data set smaller without losing any quality. Another task at this step is the data association. For workflow analysis, it is important to know the location of all tags at the same timestamp. However, all tags do not respond at the same time. Therefore, empty frames of data with timestamps at each second are created for each tag. After that, location data for each tray are associated with the workers’ location data. The outcome of this step is a data frame with some missing values. Therefore, we need one more step to finish the pre-processing, which is missing value replacement. As mentioned earlier, a UWB- based RTILTS can send the location update up to ten times within a second. However, sometimes, there can be no response in a second or more. This is caused by the inactivity of the sensor due to no movement. Therefore, it can be easily stated that if location data are missing for a given second, the location of the tag is not updated after the last recorded location data. Hence, in the missing value replacement step, the missing location data are updated according to this statement. Due to the high frequency of response from the UWB tags, the percentage of missing data are low. The missing value percentage for all tags lies between 2% ~ 5%, with an average of 2.8%. Besides, the time period of the continuous missing values varies between 1 ~ 6 seconds. Therefore, it is considered as a minor issue. This step creates a consistent data frame for each tag with location data at each second. FIG.5 shows the output of keeping one record and data association, where the data frame has some missing values. This last step shows the complete outcome of the data pre-processing unit after the missing values are filled. Movement Detection Unit (202B) The RTILTS 110 provides a continuous stream of location data for the processor 102 without any location label. Executing the movement detection unit, the processor 102 analyzes the pre-processed data to identify the location and task for different vegetable grafting 13 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 operations. In some examples, there are five different steps performed by this unit, which are: 1) location changepoint detection, 2) data association, 3) data labeling, 4) data filtering, and 5) data ordering. In most vegetable grafting operations, each task is performed by an assigned worker and takes place in a specific location. For the grafted example application of the framework 200, there is a seating arrangement for workers to perform cutting, clipping, and joining tasks. One worker is assigned to perform one task among those. Therefore, tracking the location of workers and plant trays and associating the data based on the timestamp in a synchronous manner is critical to perform the workflow analysis. In order to do that, the movement detection unit 202B starts by finding the location changepoints. The distance between the worker and a tray is used to detect the changepoints. At first, the relative distance between the workers and the tray is calculated. Since each worker requires some time to perform the assigned task, the distance between worker and tray remains the same while performing the task. However, location data from UWB tags have some noise. Sometimes this noise can mislead the change of location. The noise primarily happens due to the environmental conditions of the surrounding area. The reflection and refraction of the signal from various obstacles while transmission from tags causes the anchors to receive different location information at different times. This is also known as the multipath effect. Sometimes the signal from the tags could be entirely blocked by the obstacles, which can cause noises in location data as well. Besides, the presence of other wireless devices such as mobile phones, Bluetooth devices also causes noise in location information. Here, a binary segmentation algorithm (Scott and Knott, 1974, Sen and Srivastava, 1975) can be implemented for changepoint detection, which rules out the possibility of identifying noise as a changepoint. Binary segmentation is one of the most widely used multiple changepoint detection algorithms (Killick and Eckley, 2014). Multiple changepoint detection algorithms are extensions of single changepoint detection (SCDA). SCDA can be described as a hypothesis test where the null hypothesis Ho is no changepoint exists and alternate hypothesis H1 is one changepoint exists. The test statistic and mathematical formulation of the problem can be found in Killick and Eckley (2014). Binary segmentation first applies the SCDA to the entire data. If H_o is rejected, a changepoint exists, and the data are split into two parts at the change-point. The SCDA is then applied to these two new data sets. If new changepoints are detected in any of these new data sets, they are split again. This process is repeated until all changepoints are detected. Binary 14 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 segmentation is computationally fast as it only considers a subset of the 2 ^{^^} − 1 possible solutions. Binary segmentation can be applied to detect the changepoints in the distance data between workers and trays. For instance, for the vegetable grafting operation under study, there are four workers. After calculating the distance between each tray and all workers, changepoints are identified by applying binary segmentation. FIGS.6A-6C shows the sample output of identifying multiple changepoints using binary segmentation. It provides the row index at which distance has changed. Based on that, the mean and variance of distance data within two changepoints and other time-related statistics (i.e., start time, end time, duration) can be calculated as shown in FIG.6A. FIG.6B shows the distance points vs. time plot. Changepoints are marked with vertical dotted lines, and the mean distance for each changepoint interval is drawn with the straight horizontal lines. FIG.6B also shows that small noises in location data are not detected as a change-point. Under the proposed application of the framework 200, the parameter of the number of changepoints to be detected is set as 20. This selection is made based on the study of penalty values vs. the number of changepoints for each tray. The penalty value protects the binary segmentation algorithm from overfitting, and it decreases with the increase in the number of changepoints. We have observed that most of the trays’ penalty value becomes close to zero after 20 changepoints, as shown in FIG.6C. The selection is not optimum for all trays. However, it provides a faster solution compared to finding the optimum number of changepoints for binary segmentation. Besides, having the same number of changepoints has added benefits regarding data association in the next step, as all data sets have the same number of rows. The data association step combines the outcomes of the previous step for each tray. After that, each row of the combined data set is labeled according to the tag name and worker. The data are then filtered based on a user-defined threshold for the distance between workers and each tray. For instance, while performing any task on a tray, the tray must be within reach of the worker. In this case, it was observed that the maximum distance the worker tag can be away from the tray tag is 120 cm. Therefore, changepoints beyond this limit refer to no actual processing taking place. After filtering those changepoints, the data can be ordered based on the time. Workflow Analysis Unit (202C) 15 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 This unit aims to calculate different workflow related metrices such as different processing times, flow times, and waiting times. The unit also visualizes the result in different Gantt charts and bar charts. The steps performed by the processor 102 executing this unit are: 1) worker overlap time removal, 2) tray overlap time removal, 3) duration calculation, 4) workflow determination, and 5) plot generation. The outcome of the movement detection unit can have two types of overlap in time. One is between workers, and the other is between trays. This issue arises because the movement detection unit analyzes each pair of trays and workers independently. However, one tray cannot be processed by two workers simultaneously, and one worker cannot perform a task on two trays simultaneously. Therefore, in the proposed framework, we remove these overlap times in two steps. First, the framework tries to identify the overlap time between workers for a tray. After identifying that, the overlap removal process compares the distance between the tray and workers. The overlap time is assigned to the worker who has a lower distance compared to others. This process continues until all overlaps are removed for all trays. FIGS.7A-7C shows an example of the worker overlap time removal process. Suppose there are three workers (i.e., ^^₁, ^^₂, ^^₃), and each of them is assigned to perform a task on vegetable grafting. The ^{movement detection algorithm detects the time for performing the tasks by ^^1, ^^2, and ^^3 as} _{[ ^^11, ^^31], [ ^^21, ^^41], and [ ^^51, ^^61] respectively depending on the distance between tray and} worker (i.e., ^^₁₁, ^^₂₁, ^^₃₁) as shown in FIG.7A. There is an overlapping time period, [ ^^₂₁, ^^₃₁] between ^^₁ and ^^₂. To resolve this, distance ^^₁₁ and ^^₂₁ are compared. If ^^₁₁ > ^^₂₁, ^^₂ is closer to the tray than ^^₁. Therefore, the overlap time is assigned to ^^₂ and updated times for performing the tasks by ^^₁, ^^₂, and ^^₃ are ^[ ^^₁ ^∗ 1 , ^^₂ ^∗ 1 ^], ^[ ^^₂ ^∗ 1 , ^^₃ ^∗ 1 ^], and ^[ ^^₄ ^∗ 1 , ^^₅ ^∗ 1 ^], respectively (i.e., FIG.7B) Similarly, overlap time can be resolved in the same manner for the case when ^^₁₁ < ^^₂₁ as shown in FIG.7C. Time overlap can happen between trays as well. The tray overlap time removal step tries to solve the issue similarly to the previous step, as shown in FIGS.8A-8B. FIG.8A shows a scenario where overlap times exist between tray-1 and tray-2 after the processing in the previous step. The overlap times are assigned to the workers after comparing the distances. There are many possible scenarios to compare distances. One possible scenario can be ^^₁₁ > ^^₁₂, ^^₂₁ < ^^₂₂, ^^ ^^ ^^ ^^₃₁ > ^^₃₂, the outcome of which is shown in FIG.8B. This step takes into consideration all trays for all workers to make sure that no worker is assigned to multiple trays 16 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 simultaneously. The first two steps of the workflow analysis unit are critical to measuring different workflow metrices in subsequent steps. Besides, these steps heavily rely on the localization’s accuracy, which is why we have chosen UWB based RTILTS to provide better location tracking accuracy compared to other RTILTS technologies. The next step is to calculate the duration based on the updated timestamps after overlap time removals. This step provides the estimated processing time for each task or the estimated time the tray has spent with each worker. The flow time is also calculated. Flow time refers to the duration from the beginning of the cutting time to placing the finished grafted plant tray in the healing department. The workflow is then determined, which provides a series of durations labeled with worker and waiting times before and after each task for the length of flow time. Based on the workflow of each tray, Gantt charts for tray and worker, bar plots of processing times, flow times, and waiting times are generated. Data Collection For the example application of the framework 200, a UWB based RTILTS was implemented in a commercial vegetable grafting facility. This grafting operation is rather small (only four workers), relative to a moderately sized operation (20-100 workers); however, it served well for testing the inventive workflow methodology using a limited number of sensors. FIG.9A shows the layout of the facility and RTILTS implementation there. The facility is divided into five different departments or stations: 1) germination, 2) healing, 3) grafting, 4) office, and 5) shipping, as shown in FIG.9A. Either walls or thick curtains separate these five spaces. Therefore, a large UWB based RTILTS network setup was implemented with gateways. The anchors are gateways mounted on the walls (see FIG.9C and FIG.9D). Among different movements in the facility, we are interested in tracking the movement happening in germination, healing, and especially in grafting. For grafting, the three tables, table-A, table-B, and table-C, are set in an “I” formation (see FIG.9B). An experienced grafter set this table formation to facilitate the smooth flow from one grafting operation to another. Also, the tables are set in such a way that every workstation is at a distinct distance from the origin for better movement detection using binary segmentation. However, depending on the application, the arrangement may vary, and origin can be set accordingly. As long as the workstations are located at different distances, the variation of the spatial arrangement of workstations will not affect the accuracy of the proposed workflow analysis framework. Besides, 17 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 the RTILTS system provides the location information in cartesian coordinates ( ^^, ^^, ^^) with an accuracy of less than 10 cm (Decawave, 2019). This accuracy can be achieved by properly deploying the RTILTS 110. Therefore, different layouts may require different RTILTS network setups. However, it will not affect the accuracy of the proposed framework. Deployment of the RTILTS 110 can be done by following the methodology proposed in Chowdhury et al. (2021) to ensure the accuracy mentioned above, which is applicable to other layouts as well. Here, Table-A is the working station of the cutting worker, table-B is for clipping, and table-C is for joining. FIGS.10A-10C show the plant trays and different tasks of the grafting operation. The cutting tasks for scion and rootstocks are similar. After cutting, clips are placed at the end of the cut of rootstocks. A toothpick is also placed for the support of the plant, as shown in FIG.10B After that, the joiner places the scions inside the clips at the right orientation and size to match the rootstocks (see FIG.10C). The joining step is critical to successful grafting and requires visual inspections and grafting expertise. Therefore, this task is the most time consuming among all these three tasks. There are four workers in the example application of the framework 200: one cutter, one clipper, and two joiners. The cutting worker ⁽ ^^₁ ⁾ and clipping worker ⁽ ^^₂ ⁾ are considered ^{as novices in terms of their expert level. One joiner ( ^^3) is considered as an expert and another} _{( ^^4) is considered as intermediate. Even though ^^3 and ^^4 are assigned to perform the joining} task, sometimes they have helped ^^₁ and ^^₂ if they have no trays left for joining. However, ^^₁ and ^^₂ have never switched their roles. Besides, all workers did not start working at the same time. ^^₁ and ^^₃ started working around 9:30 AM, ^^₂ at around 10:30 AM and ^^₄ after 1:30 PM. We had twenty-four UWB modules (tracking devices 112) to generate the signals 114 and data to perform the workflow analysis. Among those, ten modules are configured as anchors and deployed to create RTILTS coverage in the facility. Three anchors were placed in the grafting department, three in the germination department, and four in the healing department. Two gateways are deployed for data collection and to support the large RTILTS network. The origin of the RTILTS network is set at one corner of the grafting department, as shown in FIG. 9A. All anchor coordinates are measured with respect to the origin and manually entered via the web interface to set up the RTILTS network. FIGS.11A-11E depict aspects of an example web 18 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 interface, which shows the list of UWB tags, anchors, and gateways, along with real-time movements. The rest of the UWB modules were configured as tags to track the workers and plant trays. In total, sixteen trays were tracked. However, after assigning four tags to the worker, we had only eight tags left to track trays. Therefore, each tag was used twice to track two different trays. After placing the grafted plant tray in the healing department, it was removed and placed in another tray to be tracked. The web interface allows the user to configure the UWB modules and observe the RTILTS network in real-time. However, it is unable to save location information. Therefore, a java application was developed for data collection purposes. The application allows saving the data into a text file in real-time, which contains the timestamps, tag identification, and ( ^^, ^^, ^^) coordinates with respect to the origin. The outcomes of analyzing the data based on the example application of the framework 200 are discussed in the next section. Results and Discussion In some examples, the proposed RTILTS based workflow analysis framework solely relies on the accuracy of the location tracking. The Decawave DWM1001 RTILTS module promises to provide a localization accuracy of less than 10 cm (Decawave, 2019). Multiple indoor experiments were conducted and confirmed that the RTILTS system could achieve that. Based on the collected location information, FIGS.12A-12B show the location plot for all tray tags. It can be observed from FIG.12A that there are some noises in the location data. After finishing the steps in data pre-processing and movement detection units, the framework 200 is able to track the movement very well. FIG.12B shows the twenty changepoints identified by the example application of the framework 200 for a tray in vegetable grafting. The tray is processed at table A (i.e., changepoints 1-3) first and then moved to table B (i.e., changepoints 4- 7) and table C (i.e., changepoints 8-9). After that, it is placed into the healing department (i.e., changepoints 10-13), and the tag is then removed and returned (i.e., changepoints 14-20) to track the next tray in the grafting department. A similar tracking movement plot can be obtained for all other tray tags. As mentioned earlier, each tag is used to track two different trays. The tray tags are named in ^^ _{^^ ^^} format where index ^^ stands for the tag number ( ^^ = 1,2, … ,8) and index ^^ stands for first tray/ second tray ( ^^ = 1,2). For instance, ^^₁₁, ^^₁₂ denote that the first tray tracked 19 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 by tag-1 is ^^₁₁ and the second tray tracked by the same tag is ^^₁₂. In total, 16 trays are tracked. Worker tags are denoted by ^^₁, ^^₂, ^^₃ and ^^₄. Depending on the RTILTS outcome shown in FIG.12B, the workflow analysis unit analyzes the vegetable grafting operation to identify some key metrices related to workflow. It also generates appropriate visualization to assess the outcome. For instance, FIG.13 shows the most commonly used workflow metrices, such as total processing time, flow time, and waiting time for each tray. These workflow metrices provide valuable insights into the overall workflow at the aggregated level. However, it does not provide direct information regarding the processing time of each task, waiting time in between different tasks, and the worker responsible for performing the task. For instance, it can be seen from FIG.13 that ^^₇₁, ^^₂₂ and ^^₁₂ have unusually longer waiting times than others, but the reasons behind these occurrences are not clear. Therefore, this information is further broken down to form the Gantt charts, which depict the overall timeline for different trays and workers. FIGS.14A-14B show two different types of Gantt charts of tray processing times to illustrate the workflow, (a) tray vs. the timeline for different tasks and (b) worker vs. the timeline for different trays. It can be observed from these two charts that there is no overlap time between trays and workers, which means our overlap time removal steps are able to resolve overlapping time segments. The Gantt chart also shows the waiting time between the tasks and how many workers were involved for each tray. From the Gantt chart of ^^₇₁, ^^₂₂ and ^^₁₂, it is found that these workers were engaged in processing other trays. However, the trays need to be processed in the order they move from one worker to another worker. This was not maintained all the time. Therefore, some trays remained unattended for a long time. Identifying such problems in vegetable grafting is critical because a longer waiting time may affect the survival rate of grafted vegetables. To have better insight regarding the workflow, we further investigate individual task time for each tray. FIG.15 shows a bar chart of different worker times for different tasks in sequence as they were performed for each tray. Each bar consists of four processing time values. If any worker did not perform any task for a tray, the processing time value is zero. For instance, ^^₁₁ was processed by only three workers (i.e., ^^₁, ^^₂, ^^₃) and ^^₈₁ was by one worker (i.e., ^^₂). Therefore, ^^₁₁ contains a zero for ^^₄ and ^^₈₁ has three zeros for ^^₁, ^^₃, and ^^₄. From Fig.15, it can be seen that most of the trays were processed by at least three workers in sequence of ^^₁ → ^^₂ → ^^₃ or ^^₁ → ^^₂ → ^^₄ which is most likely based on the assigned task for 20 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 vegetable grafting. These sequences are not followed for some trays, and for few trays, less than three workers were involved. For instance, five trays (i.e., ^^₂₁, ^^₆₁, ^^₁₂, ^^₈₁, ^^₈₂) involve less than three workers. Only the cutting task was performed on ^^₈₁. The reason is there were not enough scions to finish the joining task. The other four trays involve two workers. This is because a single worker performed multiple tasks. For instance, ^^₂₁, ^^₆₁ was completed by ^^₁ and ^^₃. Since ^^₁ was assigned to perform only the cutting task, ^^₃ had to finish both clipping and joining tasks. Similarly, ^^₄ had performed both clipping and joining tasks on ^^₁₂. For ^^₈₂, both cutting and clipping were done by ^^₃ and joining was done by ^^₁. These observations are made based on the average time each worker spent with each tray. In order to find the average time, we first identify outliers in processing time data for each worker. Outliers are identified by separating the trays first that were processed by three workers. After that, a ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ = ( ^^ − ^^)/ ^^ is calculated for each task time corresponding to each tray and each worker. Here, ^^ is the processing time for a given tray for a worker, ^^ and ^^ are the average and standard deviation of processing times for a worker (i.e., ^^ = ^̅^). Trays with the probability value of the ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^, less than 0.1 or greater than 0.9 are marked as outliers. Table 1 Processing time measurement error and accuracy. Worker W W W W 1 2 3 4 Mean Error 1. 1. 1. 1. (mins) 97 52 58 46 SD Error 0. 1. 2. 0. (mins) 65 07 03 57 Accuracy 9 9 9 8 (%) 1.60 2.40 3.30 5.00 FIGS.16A-16D show the trays that are not considered as outliers for each worker. From the box plot for each worker in FIG.16E, it is evident that there are significant differences between the processing time for different tasks and workers. In order to verify that, Tukey’s multiple comparison (Tukey, 1949) test was performed, which compares the difference between each pair of worker combinations with appropriate adjustments for multiple testing. This test is also known as Tukey’s honestly significant difference test or Tukey’s HSD, which is used to 21 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 determine which means amongst a set of means differ from the rest. FIG.16E shows the test results where different workers are grouped, which is denoted by the letters based on the average processing time and variance. From the result, it is evident that there is a significant difference in processing time between ^^₁, ^^₂, and ^^₃. ^^₁ and ^^₄ are grouped due to having similar processing times. This result supports the actual worker category mentioned earlier in the experiment section that W3 was considered as an expert. Even though ^^₃ performed the most time-consuming task (i.e., joining), the average time required is significantly less than other workers. Besides, these results also show that the workflow is not balanced, which is also a reason for having waiting times shown in FIG.15. Therefore the proposed workflow analysis framework can play a major role in identifying bottlenecks and improving overall productivity. For instance, a line balancing optimization problem can be formulated using those processing times to find out the optimal number of workers at the different efficiency levels to achieve a grafted plants production goal in a given production time. To measure the performance of the grafting application of the framework 200, we have defined the following accuracy measurement for each tray. Processing time calculation accuracy = [^{| ^^ ^^ ^^− ^^ ^^ ^^|} ^^ _{^^ ^^} ] × 100%

Where, ^^ _{^^ ^^} = Processing time calculated by the model for any given task ^^ _{^^ ^^} = Actual processing time for that task Here, the Processing time calculation accuracy considers only the total time required to finish a task. | ^^ _{^^ ^^} − ^^ _{^^ ^^}| calculates the absolute error in processing time measurement for a given task. Actual processing times were collected from video recordings of the experiments. Besides, processing times were recorded via direct observation as well. After comparing actual with RTILTS based workflow data, the above accuracy measurements are calculated for each tray. Table 1 shows the mean and standard deviation (SD) of errors in processing time for different workers. The accuracy for ^^₁, ^^₂ and ^^₃ are above 90%. For ^^₄ it is lower because fewer trays are processed by W₄ compared to other workers that contain high processing time variability, as shown in FIG.16E. 22 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 FIG.17 shows the stacked bar plot of errors (i.e., | ^^ _{^^ ^^} − ^^ _{^^ ^^}|) in time measurement for different trays and their corresponding task. The for ^^₃ at tray ^^₇₁. This tray has the largest waiting time compared to others.

labeled some time slots as processing time when ^^₃ was nearby and processing some other trays. Based on the results for all trays, the average processing time calculation accuracy is 90.5%. Discussion Summary In this disclosure, a UWB-based RTILTS framework is proposed for workflow analysis. Unlike the traditional workflow measurement method, which requires human intervention, this framework can autonomously and satisfactorily perform a workflow analysis. The framework can accurately detect movement of objects in a production process across different stations, e.g., can detect the movement of plant trays from one workstation to another. It can also detect whether a UWB tag is reused or not. The binary segmentation of the relative distance between the worker and tray can identify when the worker performs a task on a tray. Besides, the overlap removal process can successfully remove the overlap time to ensure that no worker is assigned to perform multiple tasks and no tray is processed by multiple workers simultaneously. The visualization of results in different Gantt charts and bar charts provides valuable insight regarding the workflow. It helps identify process improvement opportunities such as line balancing, layout optimization, and resource management. Results reveal that it can detect workflow with an accuracy of 90.5%. This accuracy is quite promising for the vegetable grafting workflow analysis. However, it can be further improved by applying unsupervised classification algorithms such as the gaussian mixture model or k-means clustering in conjunction with binary segmentation. The classification algorithm can label location information based on their similarity. After that, the location information of each class can be segmented using binary segmentation to identify tasks more precisely. In some examples, additional data processing can be implemented to identify the worker’s task who is not assigned to do that. Since vegetable grafting only requires movement of hands and fingers, data glove-based hand motion detection (Masoud et al., 2019b) can be used in fusion with the RTILTS information. The application of the classification algorithm and fusion of RTILTS data with data glove information are considered as opportunities for future studies. The description of the disclosure is provided to enable a person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to 23 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Throughout this disclosure the term “example” or “exemplary” indicates an example or instance and does not imply or require any preference for the noted example. Thus, the disclosure is not to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. Illustrative aspects of this disclosure include: Statement 1. A method that includes operations for real-time indoor location tracking to identify worker and task associations in a labor-intensive production process, based on location data. The method includes operations executable by a processor for accessing location data from a set of more tracking devices (of a real-time locating system (RTLS)), preprocessing the location data to identify locations of a tag associated with an object or worker over a time series, detecting changepoints via movement of an object relative to a worker and/or predefined origin, and conducting workflow analysis to compute one or more metrics based on differences between two or more of the changepoints to illuminate possible adjustments for improving the manufacturing process. Statement 2. The method of statement 1, further including conducting data separation step to separate the location data into different data frames defining different objects in the manufacturing process by identification of a threshold distance of a given tag from a predefined origin. Statement 3. The method of any one of statements 1-2, further including detecting the changepoints using binary segmentation, such that the processor: (i) applies single changepoint detection (SCDA) to detect a changepoint from the location data, (ii) segments the location data into segmented data at the changepoint, and (iii) applies the SCDA to the segmented data to identify a possible new changepoint. Statement 4. The method any one of statements 1-3, wherein the processor applies steps (i)-(iii) of statement 3 until all changepoints are detected from the location data. Statement 5. The method any one of statements 1-4, wherein the location data comprises a time series defining a location of the plurality of tags at different intervals of the time period, the location of the plurality of tags being generated in cartesian coordinates and calculated based on a distance from a predefined origin. 24 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 Statement 6. The method of any one of statements 1-5, further comprising: associating the objects with at least one worker having a location within a threshold distance from the objects at a given timestamp of the time period. Statement 7. The method of any one of statements 1-6, wherein the changepoints reflect a time spent by the objects at each of a plurality of stations associated with the manufacturing process. Statement 8. The method of any one of statements 1-7, wherein the set of tracking devices uses ultra-wideband technology to receive a continuous stream of real-time locations from tags affixed to each of the workers and each of the objects. Statement 9. A system comprising a processor configured to execute one or more processes, and memory configured to store a process executable by the processor. The process, when executed, is operable to perform operations according to any of statements 1-8. Statement 10. A non-transitory, computer-readable medium storing instructions encoded thereon. The instructions, when executed by one or more processors, cause the one or more processors to perform operations according to any of statements 1-8. Additional aspects of this disclosure are set out in the independent claims and preferred features are set out in the dependent claims. Features of one aspect may be applied to each aspect alone or in combination with other aspects. In addition, while certain operations in the claims are provided in a particular order, it is appreciated that such order is not required unless the context otherwise indicates. It should be understood from the foregoing that, while particular embodiments have been illustrated and described, various modifications can be made thereto without departing from the spirit and scope of the invention as will be apparent to those skilled in the art. Such changes and modifications are within the scope and teachings of this invention as defined in the claims appended hereto. 25 91440983.3

Claims

Attorney’s Docket No.: 085067-773789 UA20-202 CLAIMS What is claimed is: 1. A real-time tracking system for automated workflow analysis based on location tracking, comprising: a set of tracking devices including a plurality of tags configured for engagement with objects and workers of a manufacturing process, wherein the set of tracking devices generate signals indicative as to a location of the workers and objects relative to predetermined locations associated with tasks of the manufacturing process; and a processor having access to the signals and in operable communication with a memory, the memory storing instructions the processor executes to: preprocess the signals to generate location data that tracks the objects through the manufacturing process over a time period in a predetermined format, detect changepoints in the location data during the time period using at least a distance between the workers and the objects, and compute one or more workflow metrics based on differences between two or more of the changepoints to illuminate possible adjustments for improving the manufacturing process. 2. The real-time tracking system of claim 1, wherein the location data comprises a time series defining a location of the plurality of tags at different intervals of the time period, the location of the plurality of tags being generated in cartesian coordinates and calculated based on a distance from a predefined origin. 3. The real-time tracking system of claim 1, wherein the processor conducts a data separation step to separate the location data into different data frames defining different objects in the manufacturing process by identification of a threshold distance of a given tag from a predefined origin. 4. The real-time tracking system of claim 1, wherein the processor detects the changepoints using binary segmentation, such that the processor: (i) applies single changepoint 26 91440983.3 Attorney’s Docket No.: 085067-773789 UA20-202 detection (SCDA) to detect a changepoint from the location data, (ii) segments the location data into segmented data at the changepoint, and (iii) applies the SCDA to the segmented data to identify a possible new changepoint. 5. The real-time tracking system of claim 4, wherein the processor applies steps (i)-(iii) until all changepoints are detected from the location data. 6. The real-time tracking system of claim 1, wherein the signals include radio wave signals sent back and forth from a moving tag of the plurality of tags to at least one stationary anchor, the processor configured to: measure a time of flight of the time taken by the signals to make a trip between the moving tag to the at least one stationary anchor by two way ranging, and derive a value of the distance between the moving tag of the plurality of tags to the at least one stationary anchor by dividing the time of flight measured from the at least one stationary anchor by a speed of radio waves. 7. The real-time tracking system of claim 6, wherein processor utilizes the value of the distance between the moving tag and multiple anchors to calculate a location of the moving tag using triangulation. 8. The real-time tracking system of claim 1, wherein the processor associates the objects with at least one worker having a location within a threshold distance from the objects at a given timestamp of the time period. 9. The real-time tracking system of claim 1, wherein the changepoints reflect a time spent by the objects at each of a plurality of stations associated with the manufacturing process. 10. The real-time tracking system of claim 1, wherein the set of tracking devices uses ultra- wideband technology to receive a continuous stream of real-time locations from tags affixed to each of the workers and each of the objects. 27 91440983.3