System and method for robust detection of individual access to feeder / drinker using RFID LF with temporal consolidation of concurrent readings and generation of indicators and alerts in livestock farming

Abstract

The invention describes a system for obtaining individual animal access events at a feeder (4) and / or waterer (3) using low-frequency RFID readers (2) operating at 125 kHz or 134.2 kHz, compliant with ISO 11784 / 11785 where applicable, which read ear tags (1) even with multiple concurrent access. One or more readers (2) generate repeated discrete readings of the animal's identifier (1) (8), which an edge device (5) consolidates using a time-based procedure that determines the start and end of access based on no-read intervals and persistence windows. The edge device stores events in a local buffer in case of connectivity loss and subsequently synchronizes them with a remote server. The server builds time series and behavioral indicators (frequency, duration, and time distribution) and generates alerts for deviations from historical references. A web and / or mobile interface presents prioritized alerts and operational lists.

Description

System and method for robust detection of individual access to feeder / drinker using RFID LF with temporal consolidation of concurrent readings and generation of indicators and alerts in livestock farming TECHNICAL SECTOR The invention falls within the sector of digital livestock farming, industrial IoT systems, radio frequency identification (RFID) via an industrial bus, information transfer / storage via a local processing gateway, and data / time series analytics for early detection of health and management issues in intensive bovine and porcine farms (fattening, rearing), facilities where the presence of metallic elements and dirt can pose a challenge for this type of system. BACKGROUND OF THE INVENTION In intensive rearing and fattening environments, the detection of disease, stress, or poor adaptation often relies on visual observation without continuous quantitative recording, which can lead to delayed interventions and significant production losses. RFID-based monitoring solutions exist that record animal access to feeding areas and store time events in transactional / analytical systems. Specifically, US patent 11432531B2 describes systems for individual bovine identification using RFID tags, detection within a defined feeding area, and storage of entry and exit times, as well as antenna configurations to generate a read volume within the feeder. On the other hand, Hana Micron's patent family US11138864B2, US2012326862A1, and US2012326874A1, relating to early warning systems, describes an approach in which an activity measurement zone (AMZ) is defined near an incentive device (e.g., feeder / waterer), and animal activity is detected and counted using RFID to generate an alert when that activity falls below a threshold. It also describes an embodiment geared toward rural or remote deployments in which the system can be powered by alternative hybrid sources, comprising a wind turbine, solar panel, charge controller, and battery. Additionally, it discloses a variant in which the RFID antenna is integrated into a crossbar of a fence in the fattening unit to define the reading range around the feeder area and maintain the AMZ + activity count + alert threshold scheme. However, in real-world operations, significant technical challenges remain, such as robust RFID reading capture in the presence of metal, dirt, and moisture, and managing concurrent access when more than one animal enters the instrumented point simultaneously. Furthermore, it is essential that the deployment be repeatable with minimal technical support (plug & play), automatically and in real time transform simple events into actionable operational alerts, guarantee security, traceability, and multi-farm integration, and implement granular access control (roles) and operational auditing. In light of SKYSONG INNOVATIONS LLC's patent family relating to cattle monitoring using RFID technologies in feeding areas, as well as other documents on anomaly detection in time series in domains other than livestock, it is clear that the state of the art does not anticipate a solution that combines, in a single integrated system, low-frequency RFID reading at feeder and waterer access points under adverse environmental conditions and with multiple animals present, an industrial edge computing architecture (also called a perimeter gateway or border node) per pen based on RS-485 bus or wireless connectivity with local buffering capacity and cellular connectivity, and a cloud-based time series analytics backend that calculates specific intake and drinking indicators and applies adaptive thresholds based on historical data and seasonality.All of this is on a multi-farm platform with role-based access control, traceability, and APIs for integration with livestock management systems. On the other hand, document WO2024189623A1 discloses a system designed to monitor live weight and water consumption using a physical setup that integrates at least one drinking trough with a water meter, an RFID identification device, and an associated scale. The animal drinks in a configuration that allows for the simultaneous capture of weight and water consumption per individual. However, this document focuses on the physical and functional integration of instrumented points (drinking trough + flow meter + scale) and on associating the data with the animal's identifier. It does not address the specific technical problem of RFID LF reading in concurrent scenarios (multiple transponders within the reading volume) or how to consolidate reading events into a time series with deterministic and reproducible rules.In particular, WO2024189623A1 does not anticipate or suggest a deterministic temporal consolidation procedure for LF readings under concurrency, based on time windows, repetition / filtering, and event confirmation / invalidation criteria to minimize collisions and false negatives, which would allow for the reconstruction of reliable access to feeding points even when simultaneous access occurs. Unlike the above, the present invention is not based on a mechanical weighing configuration during drinking, but rather on (i) RFID readers (2) at feeder (4) / drinker (3) access points that generate repeated readings, (ii) an edge device (5) that consolidates these readings into access events through temporal filtering and provides buffering in the event of intermittent connectivity, and (iii) a backend that builds time series and infers alerts using adaptive thresholds and seasonality. In preferred embodiments, water consumption can be measured by a flow meter connected to the edge and associated with the animal's access to the drinking trough event, maintaining as a differentiating feature the pipeline of adaptive alert time series events, instead of the focus on weight. EXPLANATION OF THE INVENTION The invention proposes a comprehensive solution in three functional layers, as represented in Figure 1: the physical and edge processing layer (layer 1), the cloud backend (layer 2), and the user interface frontend layer (layer 3). This architecture enables robust monitoring of individual farm animal access (such as cattle or pigs) to feeders and waterers in intensive livestock operations, generating actionable alerts through pattern analysis. The physical layer is deployed in the livestock farm environment (pens or barns), where low-frequency RFID readers (LF, for example, 125 kHz or 134.2 kHz) are installed at the access points to linear feeders (element 4 in Figure 2) and / or waterers (element 3 in Figure 2). These readers are configured to read unique identifiers carried by the animals via RFID ear tags (element 1 in Figure 2), even under adverse conditions such as the presence of metal, humidity, or dirt. Each RFID reader (element 2 in Figure 2) interrogates the transponder when the animal enters the instrumented area and generates atomic readings that include, at least, the animal's identifier and a timestamp associated with the moment of detection. Readings from one or more readers are transmitted to the edge device (element 5 in Figure 2) associated with the pen or farm. This device, with an IP65 protection rating for industrial environments, concentrates these readings via an industrial bus (e.g., RS-485) or a wireless link. Local processing takes place at this edge device: multiple readings from the same animal are managed through filtering and aggregation to reduce noise and duplication; the direction of movement (entry / exit) is inferred from the observed sequence; and the readings are consolidated into access events that include at least the animal identifier, the start and end times, the estimated duration, and the location associated with the instrumented point.Additionally, the edge implements a transmission and buffering stage in which it attempts to send events to the backend via cellular connectivity (element 6 of Figure 2) and, in case of lack of coverage or link failure, persists them in a local buffer for later retry, ensuring operational continuity; the device can be powered by the grid or, optionally, by a solar panel with accumulator (element 7 of Figure 2), and incorporates telemetry for diagnosis and remote monitoring of the system status. The backend ingests access events transmitted from edge devices via an input service that validates their format, temporal consistency, and references to animals, reading points, and farms, normalizing the data to a common model before storage. The information is persisted in a database optimized for time series by animal, group, and farm, maintaining chronological order and enriching each event with metadata on origin, processing status, and traceability. This enables efficient queries by identifier, time interval, and location, as well as reconstructions of the access sequence for subsequent audits or analyses.On these time series, the backend executes aggregation processes and calculation of derived indicators, such as frequency of access to food and / or water, average duration and distribution of stay times, temporal patterns (hourly distribution, daily evolution or by production cycle) and measures of deviation with respect to historical references or the group, generating feature vectors that summarize the behavior of each animal or group in different observation windows. Based on these indicators, a parameterizable rules engine and / or anomaly detection models are applied, employing adaptive thresholds based on historical data, seasonality, and the criticality of each type of incident. This engine can combine conditions across multiple animals, pens, or periods, producing prioritized alerts that include the probable cause, severity level, and operational recommendations, as illustrated in the data transformation flow in Figure 3. The backend also manages the lifecycle of these alerts (generation, updating, closure, and action recording), exposing them to user channels and external systems via REST APIs. These APIs allow the platform to be integrated with livestock management systems, ERPs, and traceability solutions, including event queries, time series, active alerts, and configurations.The backend architecture is designed in a multi-tenant model, with strict logical separation by operation to prevent cross-access, and incorporates security mechanisms that include role-based access control (farmer, veterinarian, technician, integrator or cooperative), robust authentication, granular authorization following the principle of least privilege, encryption of communications and, where appropriate, of information at rest, systematic input validation, protection against common attacks, password and session policies, monitoring of anomalous events and an audit system that records relevant operations (additions, deletions, modifications, accesses and exports), associating user or role, timestamp and context of the operation to facilitate reviews and regulatory compliance. The frontend layer provides intuitive interfaces for interacting with the system, embodied in a web panel designed for the comprehensive management of farms, groups, and devices. This panel offers dashboards with aggregated metrics, trends, and detailed views by animal or group, allowing navigation through the multi-farm hierarchy, application of filters by time interval and location, and consultation of access history and associated alerts. From this web panel, it is possible to configure system operating parameters, including detection rules and thresholds, analysis time windows, and alert priorities, as well as monitor the operational status of edge devices (connectivity, firmware version, basic telemetry). When authorized, data or reports can also be exported for external analysis and traceability. Additionally, a mobile application designed for daily field operations provides users with prioritized alarms and notifications, dynamic lists of animals to be checked based on rules, pending anomalies or open incidents, and a set of real-time status indicators that reflect both recent activity at the reading points and the health of communications and the system as a whole, reducing response times and facilitating on-site intervention. This application also allows users to record actions on each alert (e.g., check performed, treatment applied, or closure due to cause) so that these actions are associated with the animal's history in the backend, and adapts the available views and options to the user's role (farmer, veterinarian, technician, integrator, or cooperative), respecting the defined permissions and access scopes.Both the web panel and the mobile application consume the APIs exposed by the backend and the rich alerts generated in the processing flow (Figure 3), being implemented as frontend software differentiated from the edge firmware and the backend software, but coordinated with them through well-defined interfaces and independently updatable in browsers and mobile devices. DESCRIPTION OF THE FIGURES Figure 1 illustrates the overall functional scheme of the system, represented by three interconnected layers arranged vertically: the edge hardware and processing layer at the bottom, the backend layer in the middle, and the user frontend layer at the top, with arrows indicating the unidirectional flow of data from the edge layer to the backend and then to the frontend. The edge hardware and processing layer represents the elements deployed on the livestock farm, such as feeders and waterers with access controlled by RFID readers, connected to an edge device that provides connectivity to the upper layer. The backend layer shows a cloud server that receives data from the lower layer, with symbolic representations of a database, a processing engine, and the generation of alerts that are directed to the frontend layer. The frontend layer represents a web panel with visualization dashboards and a mobile application that receives notifications and alerts, including lists of items to review and system status indicators. Figure 2 shows the passive RFID device (1) worn by each animal as an ear tag, which allows for its unique identification while using the facilities. Elements (2) are the RFID readers that capture this identification when the animal enters the coverage area. They are typically installed at access points to feeders and waterers to record each entry associated with feeding. In livestock identification solutions like this, the readers (2) can be integrated using wired interfaces such as RS-485 for stable communication in the field. Element (3) conceptually represents the waterers used on the farm, included to contextualize the usage scenario within the patent. Element (4) is the conceptual representation of the feeders, in this case linear feeders, as the infrastructure where animal access and, therefore, event capture occur.The IP65-rated edge device (5) is shown, responsible for collecting readings, managing system connectivity, and acting as a gateway between the physical environment and remote services. The antenna (6) illustrates the cellular connectivity between the edge device and the backend, through which events and telemetry are transmitted for storage and processing. A solar panel (7) with a battery is shown, designed to power the system when a direct power supply is unavailable or when energy independence is desired. Finally, the animals (8) wearing RFID tags, represented in the illustration as cattle, are identified as the subjects monitored by the system. Figure 3 illustrates a vertical flowchart representing the sequential data transformation from initial capture to alert generation within the system. The flow begins with an RFID reading generated by an LF reader when an ear-tagged animal enters an instrumented area, producing an atomic reading that includes the animal's identifier and a timestamp. Next, the local processing performed on the edge device is shown, where multiple readings are managed through filtering and aggregation, the direction of access (entry or exit) is inferred, and an access event is consolidated with the animal's identifier, start and end times, estimated duration, and location. Finally, the diagram represents the transmission and buffering stage at the edge, from where the event is sent to the backend, with local persistence in case of connectivity failure.In the backend, events are incorporated into a time series storage system structured by animal and pen, preserving chronological order and metadata. Indicators (access frequency, average duration, time patterns, and deviations) are calculated from this time series, and a feature vector is generated. The process continues with the application of a rules engine and anomaly detection using adaptive thresholds and historical grouping data, issuing an alert signal with priority and probable cause. Finally, the alert is delivered as an enriched notification (including animal context, incident type, and recommendation), published on the web panel and mobile application. PREFERRED EMBODIMENT OF THE INVENTION For the purposes of this application, an "access event" is defined as a record that associates an animal's RFID identifier with a specific point in time and, optionally, with a corresponding access point at a feeder and / or waterer, also including, where applicable, a duration. This event is generated when the animal accesses an instrumented point. A "time series" is defined as the sequence of events and / or derived indicators, such as frequency, duration, or time patterns, associated with an animal or group over time. An "adaptive threshold" is defined as a variation limit that is dynamically adjusted based on the animal's or group's history and seasonality, in order to reduce false alarms. An "access point" is defined as a spatial area within the reading range of an RFID reader, delimited by the arrangement of antennas at the entrance of a linear feeder and / or waterer. In a preferred embodiment, the RFID capture layer uses LF transponders (125 kHz or 134.2 kHz, in accordance with ISO 11784 / 11785 where applicable) in the form of a ring (1) in the animal's ear, with one or more RFID readers (2) at feeders (4) and / or drinkers (3) access points, configured for adverse conditions (humidity, metal, dirt) and the presence of 2 animals. In a preferred embodiment, the readers (2) connect to an edge device (5) configured to aggregate readings from multiple reading panels via an industrial bus, preferably RS-485 or wireless. The edge device performs time filtering and / or aggregation of readings to convert individual readings into access events and transmits these events to a backend via wired access connectivity, including fiber or copper, and / or cellular connectivity (6), including, but not limited to, 4G, 5G, CAT-M, NB-IoT, or NTN. The edge device implements local buffering to preserve events in the event of connectivity loss, retransmitting stored events when the connection is restored, and additionally sends telemetry and operational metrics for remote diagnostics, including metrics from other sensors on the farm connected to the device.Preferably, the edge device is housed in an outdoor-ready enclosure with IP65 protection, supporting temperatures and humidity typical of a livestock farm. In a preferred embodiment, the backend receives events related to animal behavior, validates them, and stores them in a time-series database. For each animal and / or for each defined aggregation, the backend calculates indicators that include, but are not limited to, the frequency of access to food and water per day and / or week, the average duration of access, temporal patterns such as hourly distribution and trends, and deviations from the usual behavior of the group and / or the animal.A deterministic analysis is performed on these indicators, comprising an if / then rule engine based on fixed and / or parameterizable thresholds per operation or grouping. For example, rules such as "if the daily frequency falls below X with respect to the N-day moving median and the average duration falls below Y for M consecutive windows, then generate an alarm." This analysis also incorporates time consolidation and event validation rules to reduce false positives due to partial readings or concurrency. In a preferred embodiment, this deterministic analysis further includes an adaptive threshold system calculated from historical data, including moving averages or medians, percentiles, hourly and day-of-the-week bands, and simple seasonality, so that the limits evolve over time without requiring learning models. Optionally, anomaly detection models, including clustering, regression, and / or unsupervised learning techniques, are run on the same indicators to capture patterns not explicitly encoded in rules. Also optionally, artificial intelligence techniques are applied for alarm prioritization, false alarm reduction, and action recommendations; this layer constitutes an additional enhancement not required for the system's basic operation. In a preferred embodiment, other sensors located on the farm, such as temperature, humidity, food weighing and water flow measurement sensors, connected to the edge device, report additional information of interest for analysis, generating the backend, based on the above, prioritized alerts and ingesting event subscribers, notifications and monitoring dashboards. In a preferred implementation, the platform provides a web panel for managing farms, groups, devices, and users, as well as a mobile application focused on prioritized alerts and daily lists of animals to be checked. The platform also features real-time status dashboards and a history of alerts and events accessible to authorized users. In a preferred implementation, the backend implements security, traceability, and multi-tenant operation mechanisms, including role-based access control (RBAC) with profiles such as farmer, veterinarian, technician, and cooperative / integrator; encryption of data in transit using TLS and encryption at rest to ensure system cybersecurity; and audit logs of actions, including reading, modifying, and deleting data. Preferably, logical separation by farm is implemented to allow operation with cooperatives and integrators, preventing cross-access between different farms. In a preferred embodiment, the backend exposes a REST-type application programming interface (API) for integration with external livestock management systems, including ERP and traceability systems, allowing queries of events and time series, reading of active alerts, management of device configurations, and synchronization of master data, including groups, animals, and feedlots. This preferred embodiment is described for illustrative purposes only and is not a limitation. The system comprises edge firmware, backend software, and frontend (web / mobile) software as separate code. The present preferred embodiment is described for illustrative purposes only and is not a limitation.

Claims

1) AN animal monitoring SYSTEM for an intensive livestock farm, comprising: a) at least one RFID radio frequency identification reader (2) arranged at an access point to a linear feeder (4) and / or a drinker (3) in a pen / farm, configured to read a passive RFID identifier (1) carried by each animal that accesses it, operating with low frequency LF RFID at 125kHz or 134.2kHz, in accordance with ISO 11784 / 11785 where applicable; b) an edge device (5), connected to a plurality of readers via an industrial bus and / or a wireless link, configured to: - receive multiple RFID readings from one or more readers with multiple animals present, - execute a deterministic time consolidation procedure which, from said sequences, generates access events, assigning to each event at least (i) an animal identifier and (ii) a start time and an end time,and managing concurrency of multiple animals through confirmation and invalidation rules based on time windows and thresholds; - temporarily storing access events in a local buffer in case of loss of connectivity with a remote server; and - transmitting access events to the server, when connectivity is available, through a communications network; c) a backend server, configured to: - receive access events from the edge device (5); - build time series of accesses per animal and per group of animals, comprising at least per pen; - calculate behavioral indicators comprising at least the frequency of accesses to feeder and / or drinker per period,access duration and access distribution time patterns; and - implement an anomaly detection component that applies adaptive thresholds calculated based on a history of animal indicators and / or a pen reference and seasonality, to generate prioritized health and / or management alerts; and (d) a user interface comprising at least one web application and / or one mobile application, configured to present said prioritized alerts and operational lists of animals to be checked on the farm. 2) System according to claim 1, wherein the RFID identifier corresponds to an ear tag (1) or RFID electronic device according to the country's regulatory standards. 3) System according to any of the preceding claims, wherein the RFID readers (2) are installed at the entrances of linear feeders (4) and / or at access points to drinkers (3) in a pen. 4) System according to any of the preceding claims,wherein the edge device (5) comprises an outdoor enclosure with IP65 protection rating and data connectors for an industrial bus comprising RS-485 and / or a wireless communication interface. 5) System according to any of the preceding claims, wherein the time filtering applied by the edge device (5) comprises: i. grouping consecutive readings of the same RFID identifier within a consolidation time window; ii. determining an access start time when a first reading is received after a period without readings exceeding a first time threshold; and iii. determining an access end time when no readings are received for a period exceeding a second predefined time threshold. 6) System according to any of the preceding claims, wherein the backend server calculates the adaptive thresholds using statistical measurements from the historical indicators of the pen and / or the animal.comprising (at least) means, standard deviations, and / or percentiles over sliding time windows. 7) A system according to any of the preceding claims, wherein the anomaly detection component comprises an if / then rule engine for predefined patterns and at least one unsupervised machine learning model that generates an anomaly index from the indicators. 8) A system according to any of the preceding claims, wherein the backend server exposes an application programming interface (API) that allows querying events and time series, retrieving active alerts, and configuring parameters of the edge device (5), the RFID readers (2), and / or the indicators. 9) A system according to any of the preceding claims, wherein the backend server implements logical separation by farm, multi-tenant functionality, and access control based on user roles, comprising at least farmer profiles,Veterinarian, technician, and cooperative / integrator. 10) System according to any of the preceding claims, wherein the user interface presents a prioritized daily list of animals to be checked, generated from the alerts, along with real-time status indicators. 11) System according to any of the preceding claims, wherein the edge device (5) sends telemetry and operational metrics to the backend for remote diagnosis of the status of the RFID readers (2), connectivity (6), and data quality. 12) System according to any of the preceding claims, wherein the edge device (5) is further configured to receive data from at least one flow meter associated with a drinking trough (3), comprising instantaneous flow rate and / or volume consumed, and to associate such data with the RFID identifier of the animal (1) and with a drinking trough access event.transmitting said data to the backend for its incorporation into the time series and the generation of alerts. 13) System according to any of the preceding claims, wherein the backend is configured to allow the configuration of one or more notification contacts, comprising at least a mobile phone number, email address and / or messaging channel, for sending alarms generated upon detecting deviations in the behavior of an animal and / or deviations associated with any of the sensors whose data are received and managed in the backend. 14) METHOD implemented by means of software for monitoring animals in an intensive livestock farm, comprising the steps of: a) reading, by means of one or more RFID readers (2) installed at access points to a linear feeder (4) and / or waterer (3) of a pen, passive RFID identifiers (1) carried by animals that access,wherein the reading is performed using low-frequency RFID (LF) technology in accordance with ISO 11784 / 11785 where applicable; b) sending the RFID readings to an edge device (5) and, on said edge device, running firmware that applies a deterministic time consolidation procedure to group the readings into access events comprising, for each animal, at least an identifier, a start time and an end time of access, managing the concurrency of multiple animals by means of confirmation and invalidation rules based on time windows and thresholds; c) storing the access events in a local buffer of the edge device when there is no connectivity with a backend server, and transmitting the events to the backend via a communications network when connectivity is available; d) building, on the backend, time series of access events for each animal and / or per pen, ordering the events chronologically; e) calculating,Based on time series, behavioral indicators comprising at least the frequency of access to the feeder and / or drinker per period, average duration of access, and hourly distribution patterns, said indicators constituting indirect measures of intake and / or drinking behavior based on access patterns; f) determining, in the backend, adaptive thresholds for said indicators based on a historical record of the animal and / or the pen and seasonality, and detecting deviations from said thresholds using a rule component and / or machine learning models; and g) generating prioritized alerts associated with animals with an increased probability of health and / or management issues, and presenting said alerts in a web and / or mobile interface for action. 15) Method according to claim 14,wherein the construction of the time series comprises associating each access event with pen metadata and point type, including feeder and / or drinker, to allow aggregate calculations by groupings of animals per pen. 16) Method according to any of claims 14 or 15, wherein the determination of adaptive thresholds comprises calculating, for each indicator, reference values ​​based on moving averages and standard deviations of the pen and / or animal indicators over a reference time window, and setting alert thresholds based on said reference values. 17) Method according to any of claims 14 to 16, further comprising: a) receiving, in the edge device (5), data from one or more sensors, hopper and / or tank fill level, ambient temperature, humidity,air quality and / or smoke sensors; b) transmitting said data to the backend; and c) generating alarms in the backend when deviations are detected with respect to thresholds, rules, or models associated with said sensors, presenting and / or sending said alarms to one or more configured contacts. 18) COMPUTER PROGRAM PRODUCT, comprising instructions that, when executed on one or more processors, cause the method according to any of claims 14 to 17 to be carried out, comprising: i. a first program code in the form of firmware executable by the edge device (5); ii. a second program code in the form of backend software executable by the backend server; and iii. a third program code in the form of frontend software executable by the web and / or mobile application.

Images