Real-time product yield estimation and tracking

Abstract

A computing device includes a receiver configured to receive processing data of at least one product processed at a processing plant, the data captured from a plurality of sensors disposed at the processing plant. One or more storage devices are configured to store a machine learning model and the received data. A controller is coupled to the receiver and the one or more storage devices. The controller is configured to estimate, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for the at least one product based on the received data.

Description

REAL-TIME PRODUCT YIELD ESTIMATION AND TRACKINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 675,858, filed July 26, 2024, which is incorporated by reference herein in its entirety.SUMMARY

[0002] A computing device includes a receiver configured to receive processing data of at least one product processed at a processing plant, the data captured from a plurality of sensors disposed at the processing plant. One or more storage devices are configured to store a machine learning model and the received data. A controller is coupled to the receiver and the one or more storage devices. The controller is configured to estimate, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for the at least one product based on the received data.

[0003] A method includes receiving data captured from one or more sensors disposed at a processing plant, the processing plant comprising a plurality of workstations. A machine learning model is used to estimate one or both of a real-time product yield and a real-time product processing throughput for each workstation of the plurality of workstations for at least one product processed at the processing plant based on the received data.

[0004] A system includes a plurality of sensors disposed at a processing plant. The processing plant comprises a plurality of workstations. The plurality of sensors are configured to capture data related to one or both of a product yield and a product processing throughput for at least one product processed the processing plant. One or more storage devices are configured to store the data received from the plurality of sensors and a machine learning model. A controller is coupled to the plurality of sensors and the one or more storage devices, the controller configured to receive the data from the one or more sensors and estimate one or both of a real-time product yield and a real-time product processing throughput using the machine learning model for each workstation of the plurality of workstations for the at least one product based on the received data.

[0005] The above summary is not intended to describe each embodiment or every implementation of the present disclosure. A more complete understanding will become apparent and appreciated by referring to the following detailed description and claims taken in conjunction with the accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The discussion below makes reference to the following figures, wherein the same reference number may be used to identify the similar / same component in multiple figures. The drawings are not necessarily to scale.

[0007] FIG. 1 illustrates a system for estimating and controlling yields at a processing plant according to various examples;

[0008] FIG. 2 illustrates a more detailed system for estimating and controlling yields at a processing plant according to various examples;

[0009] FIG. 3 illustrates another detailed system for estimating and controlling yields at a processing plant according to various examples;

[0010] FIG. 4 illustrates a process for estimating and controlling yields in accordance with examples described herein;

[0011] FIG. 5 illustrates a block diagram of a system and apparatus configured to perform the methods described herein.DETAILED DESCRIPTION

[0012] Estimating yields (for example, red meat yields) at processing plants may be difficult to accomplish accurately. Existing techniques utilizing depth camera may lack accuracy or cannot be scaled. Visual investigations are cumbersome, non-repeatable, and prone to bias.

[0013] Examples described herein use advanced computer vision techniques such as object detection, object tracking, and semantic segmentation to estimate red meat yield efficiency in a repeatable, precise, and real-time manner. While examples described herein involve red meat processing, it is to be understood that the techniques described may be applied to other types of products such as other types of food or meat, for example. Examples focus on chuck boning tables to deliver real-time streaming and inference from a plurality of closed-caption television (CCTV) cameras. Advanced tracking algorithms are described to solve practical tracking challenges associated with this problem.

[0014] FIG. 1 illustrates a system for estimating and controlling yields at a processing plant according to various examples. The yield estimation and control system 100 may be part of a distributed system such that different components are located in different locations. In some cases, the components are located in the same physical location and are connected via a local area network.

[0015] The controller 110 may include a processor 112 that receives various inputs and executes one or more computer programs or applications stored in memory 114. The memory 114 mayinclude computer-readable instructions or applications that, when executed, e.g., by the processor 112, cause the controller 110 to perform various calculations and / or issue commands. The processor 112 and memory 114 may together define a computing apparatus operable to process input data and generate the desired output to one or more components / devices. For example, the processor 112 may receive various input data including data from the sensors 130 to automatically control various functions of the yield estimation and control system 100. In some examples, the yield estimation and control system 100 includes one or more distributed controllers.

[0016] The controller 110 is configured to use data stored on one or more storage devices 120 (e.g., databases) to perform the estimations or other automatic or manual functions. For example, the one or more storage devices 120 store the sensor data. In some examples, the one or more storage devices 120 store other data such as historical data. The sensor data and / or the historical data may be associated with other parameters such as time, date, ambient temperatures, and conveyor belt speed, among others. The one or more storage devices may store other information about the inventory. In some cases, a single location, for example a processing plant, has multiple workstations associated with the processing plant. Different storage devices of the one or more storage devices 120 may be located in different physical locations coupled to the controller 110. The controller 110 may provide yield estimation and control using a model. According to various examples, the model may be trained via model training 140 coupled to the controller. The model may be continuously trained or may be static such that after it is trained, it does not change.

[0017] One or more sensors 130 may be coupled to the controller 110. The sensors 130 may also be described as either vision-based sensors or non-vision-based sensors. Vision-based sensors may include cameras that are capable of recording images, for example. Non-vision-based sensors may include one or more of radio frequency identification (RFID) sensors, infrared sensors, weight sensors, pressure sensors, optical sensors, motion sensors, temperature sensors, humidity sensors, and contact sensors, for example. According to various examples, other types of sensors may be embedded in devices that collect data and communicate over the internet or grant mobile access for device management. Technologies include low energy wireless, Bluetooth, near field communication (NFC), long-term evolution (LTE), ZigBee, other wireless protocols, etc. The sensors 130 are adapted to sense information about the processing plant. For example, the controller 110 may be able to use the sensors 130 to aide in an estimation of various yields or throughputs.

[0018] The controller 110 may use the processor 112 and memory 114 in various different systems. In particular, one or more processors 112 and memory 114 may be included in each different system. In some embodiments, the controller 110 may at least partially define a vision system, which may include a processor 112 and memory 114. The controller 110 may also at least partially define an automation system and / or a waste management system, which may include respective processors 112 and memory 114 separate from the processor 112 and memory 114 of the vision system.

[0019] FIG. 2. illustrates a more detailed system 200 for estimating and controlling yields at a processing plant according to some examples. The one or more sensors 230 collect information from at least one processing plant. The collected data may include videos and / or images of the product before, during, and / or after it has been processed. Movements of workers at the plant may be tracked using the sensors. For examplejoint movements of the workers may be tracked. While many of the sensor data described herein involve vision sensors, it is to be understood that other types of sensors may be used. For example, weight sensors on a conveyor belt may be used to determine a weight of the product before, during, and / or after processing. Temperature sensors may be used to track one or more ambient temperatures at the processing plant.

[0020] The captured sensor data may be stored on one or more storage devices 232. The storage devices 232 may be disposed on site at a processing plant. In some example, sensor data from multiple processing plant facilities may be stored at a single central location. A controller 210 may receive the sensor data via a network switch 211. In some examples, the controller 210 is implemented at a virtual machine at the plant.

[0021] The controller 210 includes a machine learning model 222 that is used to determine various information about operations at the processing plant based on received sensor data. For example, one or both of product yield and product processing throughput can be estimated using the model 222. Other types of information can be determined using the model such as areas where efficiency and / or safety can be potentially increased. Compound parameters may be calculated such as throughput versus yield or throughput versus a safety metric, for example.

[0022] In this example, throughout is defined as a number of meat pieces processed per unit time and may not be related to yield. Yield is an amount of meat extracted from a piece of meat. According to various examples, throughput relates to the number of pieces of meat that have been processed and may not depend on the yield

[0023] According to various examples, yield can be estimated using cameras such as RGB cameras at scale. Real time performance may be achieved using one or more of object detection, tracking, and segmentation algorithms using the model. Entropy based motion detection mayalso be used. This yield estimation may be done for each workstation at a processing plant, for example.

[0024] In the example of a meat processing plant such as a red meat processing plant, red meat yield can be defined as a ratio of bone (white) to a total number of pixels on a bone segment. This ratio may be converted to a scale. For example, the ratio can be converted to a scale of 0 to 3 automatically using the model. This yield estimation may be done on an entire bone or the bone may be split up into segments and the yield estimation may be estimated for each of the segments. The object detection and tracking algorithms may be used to determine a time spent carving each piece of meat at a workstation level, for example. Computer vision models may be used to track workers during processing. For examplejoint movements of workers may be tracked. These tracked movements may be used to determine a time and frequency of knife sharpening during a carving process. This information may be used to aide in the yield estimation or may be used to recommend changes to the processing procedure, for example.

[0025] According to some examples, the processing plant evaluation system uses machine learning to analyze and automate actions. For example, an artificial intelligence (Al) model may be used to estimate current product yields or predict future product yields at specified time periods. As used herein, an “Al model” is a mathematical algorithm implemented in a programming language that recognizes patterns from data and / or performs a task, either automatically or by learning from data in a supervised, unsupervised, semi-supervised, or selfsupervised fashion. The types of Al models may include generative models and predictive models, for example. Examples of Al model tasks include predicting current product yields, predicting future product yields, automating various plant operations to increase yields and / or throughput, and providing analysis of ways to increase yield and / or throughput future yields. An Al model is used herein synonymously with a machine learning or deep learning model and may comprise an artificial neural network.

[0026] As used herein, “deep learning” is a sub-field of machine learning that does not require expert feature engineering, but rather learns data features automatically from large quantities of data. Deep learning algorithms or models may comprise an artificial neural network having multiple hidden layers and many (e.g., thousands, millions, or billions) of learnable parameters. Example deep learning algorithms include convolutional neural networks, generative adversarial networks, recurrent neural networks, transformers, autoencoders, and deep reinforcement learning models.

[0027] The Al model may be trained via the model training component 250 using images from CCTV cameras mounted at plants, past estimated or actual yield values, past estimated or actualthroughput values, and past safety data. In some examples, the Al model may be trained using data collected from more than one processing plant. In some examples, the Al model is trained using data from other similar processing plants. The Al model may be used to update estimated yield data, automate various actions such as changing a speed of a conveyor belt, and to change one or more predefined thresholds associated with a trigger of the one or more automated actions.

[0028] The model training component 250 may include buckets 252 for storing objects related to the training process. Trained models may be provided to the bucket 252 via notebook 256 and annotated data may be provided to the bucket 252. The ground truth labelling component 254 may provide annotated data to the bucket 252 and the notebook component 256. The annotations may be at least partially performed automatically without input from a user. In some examples, the annotations are performed automatically except in a case where a quality threshold is not achieved. In these examples, the image may be discarded and not used in training or may be flagged for annotation by a user. In some examples, the annotations may be performed entirely automatically without user input.

[0029] An application programming interface (API) 224 is coupled to the model 222. The API may be coupled to a user interface 226 that allows for updating or viewing the data or analysis results via a graphical user interface (GUI) on the plant computer 240, for example.

[0030] The controller 210 may produce trend data based on historical yield data, for example. The trend data may be displayed in a GUI in various ways. For example, the trend data may be displayed in graphical format showing product yields or throughput over time. The GUI may display yields based on a time period. For example, the GUI may display a yield based on a time of day (e.g., night shift versus day shift) or a time of year. The controller 210 may use the trend data to compare yields between processing plants, between plants using different feedyards and / or processing different types of cattle. In some examples, yield data may be trended based on different type of muscle and marbling in cattle or on cattle diet. In some cases, the trend data is used to compare product yields between seasons, weather events, ambient temperatures, or other types of events. The trend data may be used to correlate safety incidents with one or both of throughput and product yield.

[0031] The controller 210 may take various actions based on the trend data. The controller 210 uses the trend data to predict data about future yields and / or predict future yields and / or throughputs within a particular time period. For example, the controller 210 may predict future yields or throughputs based on a time of year, day, or month. The controller 210 may predict future yields based on other types of events such as holidays and weather events.

[0032] The controller 210 may automatically perform automatic actions based on the trend data and display the trended information. For example, the controller 210 may automatically change a speed of one or more conveyor belts within the processing plant based on the trend data in a closed-loop fashion. The controller 210 may automatically issue an alert based on the trend data. For example, the controller 210 may issue an alert when it is predicted that a throughput or a yield drops below a predefined threshold. In some examples, the controller 210 may issue an alert when a safety level drops below a safety threshold. In some examples, the controller 210 issues an alert in advance of an event in which the estimated yield or the throughput is estimated to drop below respective predefined thresholds. In some examples, the controller 210 is configured to recommend certain actions based on the sensor data. For example, the controller 210 may be configured to automatically recommend worker training or procedure changes based on the sensor data.

[0033] According to various examples, the various types of data that are used to estimate product yields may be determined based on data collection using more than one method. The collected data includes sensor data (vision and / or non-vision sensor data) and manual data that is manually input by a user, for example. The data analysis step done by the controller 210 may use an averaging mechanism (e.g., median and / or mean aggregation) to determine estimated yields and / or throughputs. In some implementations, a more sophisticated information infusion method such as Dempster-Shafer rule theory can be utilized to merge the estimations from multiple approaches. The result of the merged estimations results in the final estimated product yield in this example.

[0034] FIG. 3. illustrates another detailed system 300 for estimating and controlling yields at a processing plant according to some examples. In this example, multiple abstraction layers are shown. A computer vision layer 310 may provide on-site sensors 312 at the local plant, compute, analytics, and data storage solutions that are used to perform real-time read meat yield estimation and contact time calculations. Contact Time refers to amount of time spent by a worker on a particular piece of a product (for example, meat). Therefore, higher contact time may indicate a lower throughput and vice versa. While throughput and contact time may be referred to herein, it is to be understood that one or both of contact time and throughput can be estimated and tracked. For example, in the context of chuck boning, contact time refers to time spent by a worker removing meat from one piece of chuck meat. In this example, the plurality of on-site sensors are cameras, but it is to be understood that non-vision based or other types of vision-based sensors may be used. The on-site cameras 312 are configured to capture data at the processing plant regarding the processing of the product.

[0035] The computer vision layer 310 may also include an edge computing component 315 that provides a local plant execution environment that is used to deploy and execute computer vision models 317 for red meat yield estimation and contact time calculations. The edge computing component 315 may also include a database 319 (for example, a relational database such as a SQL database) for storing sensor data. In this example, the on-site cameras 312 provide a live stream video feed to the edge computing component 315. The edge computing component 315 is located physically close to the on-site camera 312. For example, the edge computing component 315 may be located at the local plant.

[0036] According to various examples, a data pre-processor may be disposed in the computer vision layer 310 and may be used to determine images or videos that are not useful for the analysis of yield and contact time. Not useful images or videos may include substantially duplicate videos, poor quality videos that are not in an area of interest. In some examples, an operator may assist the data pre-processor in the determination of which videos are not useful. For example, the data pre-processor may flag videos or groups of videos that may not be useful and an operator makes the final decision of whether to send the videos to be used in the yield estimation. In some examples, the image pre-processor operates completely independently from the operator.

[0037] A smart bus layer 320 provides an implementation of digital solutions that provide user interaction based on outputs from the computer vision model. The smart bus layer 320 includes a front-end application 322 that is disposed at the local plant. The front-end application may provide formatted data received from the computer vision layer 310 to a user interface, for example, via a performance dashboard application 324. The data received from the computer vision layer 310 may include the estimated red meat yield and processing contact time.

[0038] The smart bus layer may also use an existing digital solutions component 325 to enhance the yield and contact time estimations. These existing digital solutions may include labor forecasting 326, demand forecasting 327, and a plant digital twin 328.

[0039] Labor forecasting models may be used to predict no shows and absences in advance. This information may be helpful for supervisors so that they can have good visibility into their labor availability in advance to plan shifts accordingly. This may be particularly relevant in worker roles that may be important for plant operation. These absences can create bottlenecks for the plant to run efficiently. One of the inputs to the forecasting model may be the skills or roles of the operators in the plant as a proxy of efficiency. The computer vision model can be used to give an accurate measurement of the efficiency of the operators through calculation of contact time or throughput.

[0040] Demand forecasting also may be estimated and tracked. This includes demand and orders from customers. Demand forecasting can help define a production schedule for a period of time such as a day, a week, or a time of year, for example.

[0041] A plant digital twin can be used to simulate plant operation. The digital twin can also benefit from the computer vision model-based time studies. The digital twin allows for running entire plant as a simulation to identify bottlenecks in plant operation based on labor and demand. The digital twin can be used to streamline processes and identify areas to invest capital to improve the throughput. One or more of labor, demand, and computer vision model outputs may be used to feed the digital twin. The digital twin can then provide outputs that can recommend changes to improve plant operation as described in more detail herein.

[0042] These existing digital solutions can be used in combination with the outputs from the computer vision model to provide advanced analytics and forecasting for plant operations. The existing digital solutions component 325 may provide correlations between yield scores calculated via the model and the existing digital solutions.

[0043] A channel layer 330 provides devices and applications for interaction with users. A plant user interface 332 may provide users at the plant access to the analyzed data. For example, the plant user interface 332 may provide past, real-time, and or forecasted yields and contact times. This data may be provided at a workstation level or at a plant level. According to various examples, the plant user interface may include multiple user interfaces located at different locations of the plant floor. For example, the user interfaces may be touch screens located around the plant floor at touch screen kiosks.

[0044] The channel layer 330 may also include a multi-modal analytics 334 component. This multi-modal analytics component may be disposed at the physical plant or may be disposed at a central location that oversees multiple plants. The multi model analytics may provide advanced data analytics via PowerBl or Tableau, for example. A user of the system may be able to view and request further analytics at a larger scale including multiple plants or business units. Data from different plants can be combined to optimize scheduling and planning at a large scale instead of at individual sites. The central control tower can allow the plants to compare their performance at the same time and could be used to gamify the entire process of optimization.

[0045] A user layer 340 may define user roles such that visualization and analytics outputs are tailor-made for different personals such as operators, shift supervisors, Managers, and business leaders. For example, business leaders and general Managers may receive high level stats such as weekly, monthly, and yearly statistics. Shift supervisors may additionally receive day to day yields. Finally, operators may receive near real time metrics on current performance. A userinterface at the plant or at a central location may allow for the assignment of these different roles. These roles may allow for different data and / or GUI functionality permissions. For example, trend data and analytics may only be provided for higher roles (e.g., business leaders and managers) whereas day to day yields or near real time metrics on current performance for a particular user may be provided for lower roles (e.g., operators and shift supervisors) as well as higher roles.

[0046] FIG. 4 illustrates a process for estimating yields and / or throughputs at a processing plant using a machine learning model according to various examples. The processing plant is configured to process at least one product. For example, the processing plant may process meat products such as red meat products. Data is received 410 via sensors at a processing plant. In an example, at least some of the sensors are cameras configured to capture one or both of video and still images at the processing plant.

[0047] One or both of a product yield and a product throughput is estimated 420 using a machine learning model and the received sensor data. Estimating the product yield may include determining a ratio of bone to a total number of pixels using at least one image. The machine learning model may be configured to estimate the real-time product yield using the machine learning model by performing one or more of object detection, object tracking, object segmentation, and entropy-based motion detection.

[0048] One or both of the product yield and the product throughput may be trended over time to create trend data. An alert may be automatically issued based on the estimated data or the trend data. For example, an alert may be issued if a yield or throughput is projected to or drops below a threshold.

[0049] In some examples, the machine learning model may be used to calculate a past, current, or projected safety metric and an alert may be issued based on the calculated safety metric. In some examples at least one action may be automated based on results of the estimation or the trend data. For example, a speed of at least one conveyor belt may be modified based on at least one of the estimated product yield, product throughput, and the safety metric.

[0050] The methods and processes described above can be implemented on computer hardware, e.g., workstations, servers. In FIG. 5, a block diagram shows a system and computing apparatus 500 that may be used to implement methods according to an example embodiment (e.g., as a computer, a mobile device, a server, a smart sensor, a control system, etc.). The components may be implemented as integrated circuits (ICs), portions thereof, discrete electronic devices, or other modules, instruction sets, programmable logic or algorithms, hardware, hardwareaccelerators, software, firmware, or a combination thereof, or as components otherwise incorporated within a chassis of a larger system.

[0051] One or more locations 504, 505, 506 that may include different apparatuses used within the system. For example, the locations 504, 505, 506 may include various electrical and / or mechanical components of a self-contained system. The locations 504, 505, 506 may include multiple workstations within a single processing plant. The locations 504, 505, 506 may be physically remote from one another. For example, the locations 504, 505, 506 may represent multiple disparate processing plants.

[0052] Each of the locations 504, 505, 506 is associated with sensors 512, 513, 514 with associated controllers 550, 552, 554 that provide data that can be fed into a controller 520. Other sources of data 518 may also be used as inputs to the controller 520, such as ambient temperatures (e.g., weather data), operational schedules, etc.

[0053] The controller 520 may include conventional computing hardware such as a central processor 521, memory 522, input / output (VO) interfaces 523, and a non-volatile data storage unit 524 (e.g., hard disk drives, solid state drives). The processor 521 may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and / or equivalent discrete or integrated logic circuitry. In some embodiments, the processor 521 may include multiple components, such as any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, and / or one or more FPGAs, as well as other discrete or integrated logic circuitry. The functions attributed to the controller 520 and / or processor 521 herein may be embodied as software, firmware, hardware, or any combination of these. Certain functionality of the controller 520 may also be performed in the cloud or other distributed computing systems operably connected to the processor 521. It is to be understood that the computing devices described herein may be a set of computing devices that are communicatively coupled via a cloud-based system, for example. For example, controller 520 can be a system of multiple controllers that operate together in a cloud-based system.

[0054] The memory 522 may include any volatile, non-volatile, magnetic, optical, and / or electrical media, such as a random-access memory (RAM), read-only memory (ROM), nonvolatile RAM (NVRAM), electrically erasable programmable ROM (EEPROM), flash memory, and / or any other digital media. While shown as both being incorporated into the controller 520, the memory 522 and the processor 521 could be contained in separate modules.

[0055] The controller 520 includes an external data interface 526 that receives data from the sensors 512, 513, 514 and produces outputs that can be acted on via a user interface 525 thatcommunicates inventory data to the user. The data storage unit 524 stores a machine learning model 528 that predicts various information about one or more of current and future yield, throughput, and safety data of each location based on the historical and present sensor and user input data. According to various examples, a trend data model 530 may be used to trend past data of one or more of the locations 504, 505, 506 with or without input from the machine learning model 528. An automation system 534 is used to perform one or more automatic actions such as automatic alerts and automatic control of conveyor belts, for example.ILLUSTRATIVE EXAMPLES

[0056] Example Exl : A computing device, comprising: a receiver configured to receive processing data of at least one product processed at a processing plant, the data captured from a plurality of sensors disposed at the processing plant; one or more storage devices configured to store a machine learning model and the received data; and a controller coupled to the receiver and the one or more storage devices, the controller configured to estimate, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for the at least one product based on the received data.

[0057] Example Ex2: The computing device of Ex2, wherein the plurality of sensors comprise a plurality of vision-based sensors configured to capture one or both of images and videos inside of the processing plant.

[0058] Example Ex3: The computing device of Exl or Ex2, wherein the at least one product comprises red meat.

[0059] Example Ex4: The computing device of any of Exl-Ex3, wherein the at least one product comprises a meat product and wherein estimating one or both of the real-time product yield and the real-time product processing throughput comprises determining a ratio of bone to a total number of pixels using at least one image.

[0060] Example Ex5: The computing device of any of Exl-Ex4, wherein the controller is configured to modify a speed of at least one conveyor belt at the processing plant based on one or both of the real-time product yield and the real-time product processing throughput.

[0061] Example Ex6: The computing device of any of Exl-Ex5, wherein the machine learning model is configured to estimate one or both of the real-time product yield and the real-timeproduct processing throughput using the machine learning model by performing one or more of object detection, object tracking, object segmentation, and entropy -based motion detection.

[0062] Example Ex7: The computing device of any of Exl-Ex6, wherein, the controller is configured to track worker movements based on the received data from one or more workers and the machine learning model.

[0063] Example Ex8: The computing device of Ex7, wherein tracking worker movements comprises tracking worker joints while on a floor of the processing plant.

[0064] Example Ex9: The computing device of any of Exl-Ex8, wherein the controller is configured to track one or both of frequency and time of worker knife sharpening.

[0065] Example ExlO: The computing device of any of Exl-Ex9, wherein the controller is configured to trend one or both of the real-time product yield and the real-time product processing throughput estimations over time to create trend data.

[0066] Example Exl 1 : The computing device of ExlO, wherein the controller is configured to automatically issue an alert based on the trend data.

[0067] Example Exl2: The computing device of any of Exl -Exl 1, wherein the processing plant comprises a plurality of workstations and the controller is configured to estimate one or both of the real-time product yield and the real-time product processing throughput at each workstation of the plurality of workstations.

[0068] Example Exl3: The computing device of any of Exl-Exl2, wherein the controller is configured to predict one or both of a future product yield and future product processing throughput.

[0069] Example Exl4: The computing device of any of Exl-Exl3, wherein the controller is configured to train the machine learning model using one or both of past product yield and past product processing throughput data.

[0070] Example Exl5: The computing device of any of Exl-Exl4, wherein the processing plant comprises a plurality of workstations and the controller is configured to estimate one or both of the real-time product yield and the real-time product processing throughput at each workstation of the plurality of workstations.

[0071] Example Exl6: A method, comprising: receiving data captured from one or more sensors disposed at a processing plant, the processing plant comprising a plurality of workstations; and estimating, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for each workstation of the plurality of workstations for at least one product processed at the processing plant based on the received data.

[0072] Example Exl7: The method of Exl6, further comprising modifying a speed of at least one conveyor belt at the processing plant based on one or both of the real-time product yield and the real-time product processing throughput.

[0073] Example Exl8: The method of Exl6 or Exl7, further comprising training the machine learning model using one or both of past product yield and past product processing throughput data.

[0074] Example Exl9: The method of any of Exl6-Exl8, further comprising trending one or both of the real-time product yield and the real-time product processing throughput estimations over time to create trend data.

[0075] Example Ex20: A system comprising: a plurality of sensors disposed at a processing plant comprising a plurality of workstations, the plurality of sensors configured to capture data related to one or both of a product yield and a product processing throughput for at least one product processed the processing plant; one or more storage devices configured to store the data received from the plurality of sensors and a machine learning model; and a controller coupled to the plurality of sensors and the one or more storage devices, the controller configured to: receive the data from the one or more sensors; and estimate one or both of a real-time product yield and a real-time product processing throughput using the machine learning model for each workstation of the plurality of workstations for the at least one product based on the received data.

[0076] Unless otherwise indicated, all numbers expressing feature sizes, amounts, and physical properties used in the specification and claims are to be understood as being modified in all instances by the term “about.” Accordingly, unless indicated to the contrary, the numerical parameters set forth in the foregoing specification and attached claims are approximations that can vary depending upon the desired properties sought to be obtained by those skilled in the art utilizing the teachings disclosed herein. The use of numerical ranges by endpoints includes all numbers within that range (e.g., 1 to 5 includes 1, 1.5, 2, 2.75, 3, 3.80, 4, and 5) and any range within that range.

[0077] The various embodiments described above may be implemented using circuitry, firmware, and / or software modules that interact to provide particular results. One of skill in the arts can readily implement such described functionality, either at a modular level or as a whole,using knowledge generally known in the art. For example, the flowcharts and control diagrams illustrated herein may be used to create computer-readable instructions / code for execution by a processor. Such instructions may be stored on a non-transitory computer-readable medium and transferred to the processor for execution as is known in the art. The structures and procedures shown above are only a representative example of embodiments that can be used to provide the functions described hereinabove.

[0078] The foregoing description of the example embodiments has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the embodiments to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. Any or all features of the disclosed embodiments can be applied individually or in any combination are not meant to be limiting, but purely illustrative. It is intended that the scope of the invention be limited not with this detailed description, but rather determined by the claims appended hereto.

Claims

CLAIMSWhat is claimed is:

1. A computing device, comprising: a receiver configured to receive processing data of at least one product processed at a processing plant, the data captured from a plurality of sensors disposed at the processing plant; one or more storage devices configured to store a machine learning model and the received data; and a controller coupled to the receiver and the one or more storage devices, the controller configured to estimate, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for the at least one product based on the received data.

2. The computing device of claim 1, wherein the plurality of sensors comprise a plurality of vision-based sensors configured to capture one or both of images and videos inside of the processing plant.

3. The computing device of claim 1, wherein the at least one product comprises red meat.

4. The computing device of claim 1, wherein the at least one product comprises a meat product and wherein estimating one or both of the real-time product yield and the real-time product processing throughput comprises determining a ratio of bone to a total number of pixels using at least one image.

5. The computing device of claim 1, wherein the controller is configured to modify a speed of at least one conveyor belt at the processing plant based on one or both of the real-time product yield and the real-time product processing throughput.

6. The computing device of claim 1, wherein the machine learning model is configured to estimate one or both of the real-time product yield and the real-time product processing throughput using the machine learning model by performing one or more of object detection, object tracking, object segmentation, and entropy -based motion detection.

7. The computing device of claim 1, wherein, the controller is configured to track worker movements based on the received data from one or more workers and the machine learning model.

8. The computing device of claim 7, wherein tracking worker movements comprises tracking worker joints while on a floor of the processing plant.

9. The computing device of claim 1, wherein the controller is configured to track one or both of frequency and time of worker knife sharpening.

10. The computing device of claim 1, wherein the controller is configured to trend one or both of the real-time product yield and the real-time product processing throughput estimations over time to create trend data.

11. The computing device of claim 10, wherein the controller is configured to automatically issue an alert based on the trend data.

12. The computing device of claim 1, wherein the processing plant comprises a plurality of workstations and the controller is configured to estimate one or both of the real-time product yield and the real-time product processing throughput at each workstation of the plurality of workstations.

13. The computing device of claim 1, wherein the controller is configured to predict one or both of a future product yield and future product processing throughput.

14. The computing device of claim 1, wherein the controller is configured to train the machine learning model using one or both of past product yield and past product processing throughput data.

15. The computing device of claim 1, wherein the processing plant comprises a plurality of workstations and the controller is configured to estimate one or both of the real-time product yield and the real-time product processing throughput at each workstation of the plurality of workstations.

16. A method, comprising: receiving data captured from one or more sensors disposed at a processing plant, the processing plant comprising a plurality of workstations; and estimating, using a machine learning model, one or both of a real-time product yield and a real-time product processing throughput for each workstation of the plurality of workstations for at least one product processed at the processing plant based on the received data.

17. The method of claim 16, further comprising modifying a speed of at least one conveyor belt at the processing plant based on one or both of the real-time product yield and the real-time product processing throughput.

18. The method of claim 16, further comprising training the machine learning model using one or both of past product yield and past product processing throughput data.

19. The method of claim 16, further comprising trending one or both of the real-time product yield and the real-time product processing throughput estimations over time to create trend data.

20. A system comprising: a plurality of sensors disposed at a processing plant comprising a plurality of workstations, the plurality of sensors configured to capture data related to one or both of a product yield and a product processing throughput for at least one product processed the processing plant; one or more storage devices configured to store the data received from the plurality of sensors and a machine learning model; and a controller coupled to the plurality of sensors and the one or more storage devices, the controller configured to: receive the data from the one or more sensors; and estimate one or both of a real-time product yield and a real-time product processing throughput using the machine learning model for each workstation of the plurality of workstations for the at least one product based on the received data.

Images